Surface mine surrounding rock blasting crack range control method

Abstract

The invention discloses a surface mine surrounding rock blasting crack range control method which comprises the following steps: determining a target blasting crack range according to geological conditions and mining requirements of a target surface mine; a plurality of preset blasting holes are formed in the surrounding rock, and the depth and the interval of the blasting holes are optimally designed according to the target blasting crack range; each blasting hole is filled with a preset amount of explosive, a plurality of segmented detonating devices are arranged, and the detonating time of the segmented detonating devices is controlled according to the target blasting crack range; explosives in the blasting holes are sequentially detonated through a segmented detonating device, and a preset blasting crack range is formed; monitoring the crack extension condition in the blasting process in real time by using monitoring equipment, and dynamically adjusting subsequent blasting parameters according to a monitoring result; according to the method, the blasting crack range is accurately predicted through the neural network, then the blasting parameters are accurately controlled, and the method has important practical significance for improving the blasting efficiency, guaranteeing the surrounding rock stability and reducing the mining cost.

Description

Technical Field

[0001] The invention relates to the technical field of open-pit mine surrounding rock blasting gap range control, and in particular to an open-pit mine surrounding rock blasting gap range control method. Background Art

[0002] In the process of open-pit mining, rock blasting is one of the key technologies to achieve efficient mining and ensure slope stability. However, traditional blasting methods have some limitations, such as the difficulty in accurately controlling the scope of the blasting gap, which leads to damage to the integrity of the surrounding rock of the slope, stress concentration, and difficulty in subsequent support.

[0003] In recent years, with the in-depth research and technological innovation of blasting technology, some new methods and technologies have been proposed to solve these problems. For example, pre-splitting blasting and smooth blasting technology effectively reduce the damage to the slope rock mass caused by blasting by forming pre-cracks in the surrounding rock. In addition, by optimizing blasting parameters (such as hole spacing, row spacing, unit consumption of explosives, etc.), the blasting effect can be further improved and the damage to the surrounding rock can be reduced.

[0004] However, although these technologies have improved blasting efficiency and surrounding rock stability to a certain extent, there are still some challenges in practical application. For example, how to achieve precise control of the blasting fracture range under complex geological conditions (such as surrounding rocks with developed joints and fissures) is still an urgent problem to be solved. Summary of the invention

[0005] The object of the present invention is to provide a method for controlling the blasting crack range of surrounding rock in an open-pit mine, so as to solve the above-mentioned problems existing in the prior art.

[0006] The specific application is as follows:

[0007] A method for controlling the blasting crack range of surrounding rock in an open-pit mine comprises the following steps:

[0008] S1. Determine the target blasting gap range according to the geological conditions and mining requirements of the target open-pit mine;

[0009] S2. Arranging a plurality of preset blasting holes in the surrounding rock, wherein the depth and spacing of the blasting holes are optimized according to the target blasting gap range;

[0010] S3, loading a preset amount of explosives into each blasting hole, and setting a plurality of segmented detonating devices, wherein the detonation time of the segmented detonating devices is controlled according to the target blasting gap range;

[0011] S4, sequentially detonating the explosives in the blasting holes by means of a segmented detonation device to form a preset blasting gap range, wherein the blasting gap range is within a preset error range;

[0012] S5. Use monitoring equipment to monitor the crack expansion during the blasting process in real time, and dynamically adjust subsequent blasting parameters based on the monitoring results.

[0013] Furthermore, determining the target blasting gap range in S1 includes:

[0014] Determine the target blasting fracture range through numerical simulation and comprehensive analysis of data from blasting operations based on the geological conditions of the target open-pit mine, wherein the geological conditions include at least rock type, fracture development degree and joint distribution;

[0015] The specific implementation of the target burst gap range includes:

[0016] S11. Geological condition analysis: Based on the geological conditions of the target open-pit mine, generate ore body structure characteristics and rock characteristics through geological exploration;

[0017] S12, numerical simulation, using a numerical simulation method to perform N blasting numerical simulations, the input of the numerical simulation method is the geological conditions and blasting parameters of the target open-pit mine, and the output of the numerical simulation method is the simulated target blasting gap range;

[0018] S13, neural network algorithm analysis, analyzing the target blasting gap range of the target open-pit mine through the trained RBF neural network, the input of the RBF neural network is the drilling parameters, rock properties, and blasting parameters of the target open-pit mine, and the output of the RBF neural network is the RBF neural network target blasting gap range;

[0019] S14. Obtaining the target blasting gap range, calculating the first similarity between the target blasting gap range of the RBF neural network and the simulated target blasting gap range generated by N blasting numerical simulations by the cosine similarity algorithm, and presetting a similarity threshold. If the first similarity between the simulated target blasting gap range output by the nth blasting numerical simulation and the target blasting gap range of the RBF neural network meets the preset similarity threshold, then finding the simulated target blasting gap range output by the corresponding nth blasting numerical simulation as the final target blasting gap range, and at the same time finding the input blasting parameters of the corresponding nth blasting numerical simulation as the initial blasting parameters.

[0020] The specific process of obtaining the degree of crack development and joint distribution is as follows:

[0021] S15, graying and binary image statistics are performed on the crack development degree image and joint distribution image obtained by the high-definition camera, wavelet transform series are selected, and secondary Daubechies wavelet transform is performed on the binary image statistics; one-dimensional Daubechies wavelet transform is used for processing;

[0022] S16, selecting a Gaussian filter and a filter radius to perform smoothing on the transformed image data, performing peak impact detection on the smoothed image data to obtain a peak distribution array, selecting model parameters, and using a peak impact response model to calculate an impact response of the peak distribution array;

[0023] S17, using the peak impulse response result, obtain the position when the peak distribution array response drops to 0 for the first time, obtain the most recent maximum value based on the response position, use it as the threshold, perform Daubechies wavelet inverse transform, restore it to the position of the original binary image, and use this position as the segmentation threshold of the original grayscale image to segment the image, and obtain data information corresponding to the degree of crack development and joint distribution respectively;

[0024] S18. Processing the data information corresponding to the degree of crack development and the distribution of joints through image morphology to obtain the features corresponding to the degree of crack development and the distribution of joints, respectively, wherein the features include the number of crack endpoints, the number of intersections, the surface crack rate and the fractal dimension value.

[0025] Furthermore, the selection of the neuron parameters of the input layer, output layer and hidden layer of the RBF neural network in S13 is as follows:

[0026] Input layer: drilling speed, rotation speed, drilling impact pressure, rotation pressure, drill bit acceleration, drill bit torque, drill rod thrust, drill bit vibration state, drilling angle, drilling depth, rock type, blasting parameters, blasting effect, so the input layer has a total of 13 parameters and 13 neurons;

[0027] Output layer: predicts the target burst gap range, so the output layer has 1 parameter and 1 neuron;

[0028] Hidden layer: The number of hidden layer nodes is adjusted according to the network training results.

[0029] Furthermore, in S5, the crack expansion during the blasting process is monitored in real time using monitoring equipment, including:

[0030] S54, using a distributed optical fiber sensor to monitor the temperature change inside the surrounding rock, inverting the expansion of the crack through the temperature change, and generating a first temperature change feature;

[0031] S55. Use optical fiber sensors and strain gauges to monitor strain changes on the surface of the surrounding rock, obtain dynamic information on crack expansion in real time, and generate a first dynamic information feature.

[0032] Furthermore, in S5, using monitoring equipment to monitor the crack expansion during the blasting process in real time also includes:

[0033] S56, using an acoustic emission monitoring sensor to monitor the acoustic emission signal of the surrounding rock in real time during the blasting process, determining the expansion position and rate of the crack through signal analysis, and generating a first signal feature;

[0034] S57. Use a multi-physical field coupling monitoring sensor to simultaneously monitor the stress, strain, temperature and acoustic emission signals of the surrounding rock to generate a first coupling feature.

[0035] Furthermore, in S5, dynamically adjusting subsequent blasting parameters according to the monitoring results includes:

[0036] S61. According to the real-time monitoring of the crack expansion, the amount of explosives, detonation time and hole spacing of subsequent blasting holes are adjusted to ensure that the crack range meets the target requirements;

[0037] S62, by establishing a mathematical model between blasting parameters and crack range, using a first blasting parameter analysis algorithm to analyze the first temperature change feature, the first dynamic information feature, the first signal feature and the first coupling feature, and predicting the optimal blasting parameters;

[0038] S63. After each blasting, the blasting effect is evaluated, and the parameters of subsequent blasting are adjusted according to the evaluation results to form a closed-loop control.

[0039] Furthermore, in S63, after each blasting, the blasting effect is evaluated, and the parameters of subsequent blasting are adjusted according to the evaluation results, so as to form a closed-loop control including:

[0040] S64, using a drone equipped with a high-resolution camera to quickly scan the surface of the surrounding rock after blasting, obtain three-dimensional distribution information of cracks, and generate a first three-dimensional image feature;

[0041] S65, using a first deep learning algorithm to perform real-time analysis on the first temperature change feature, the first dynamic information feature, the first signal feature, the first coupling feature, and the first three-dimensional image feature, identify key features of crack extension, and predict a final range of the crack; the first temperature change feature, the first dynamic information feature, the first signal feature, the first coupling feature, and the first three-dimensional image feature are divided into a training set and a test set, and the training set and the test set are used for training the first deep learning algorithm;

[0042] S66. Automatically adjust the parameters of subsequent blasting according to the prediction results.

[0043] Furthermore, the first deep learning algorithm includes:

[0044] The first deep learning algorithm uses a convolutional neural network (CNN) combined with a long short-term memory (LSTM) network; the training set is analyzed by a CNN-LSTM model to obtain a final model after the CNN-LSTM model is trained, and the convolutional neural network (CNN) uses a DenseNet neural network framework; the long short-term memory (LSTM) network is used to capture the dependency relationship between the accumulation of the first temperature change feature, the first dynamic information feature, the first signal feature, the first coupling feature, and the first three-dimensional image feature and time, and then output the final range value of the crack;

[0045] The specific implementation process of the CNN-LSTM model is as follows:

[0046] The CNN-LSTM model architecture and processing process are as follows: the CNN model and the LSTM model are connected in series, the first three-dimensional image feature is input into the CNN model, the first CNN data set is output through CNN model feature processing, and the first temperature change feature, the first dynamic information feature, the first signal feature, and the first coupling feature are input into the LSTM model;

[0047] The processing process of the CNN model is:

[0048] Applying kernel K-means clustering analysis with adaptive weight allocation to the first three-dimensional image feature Y, feature enhancement and dimensionality reduction are performed, and the expression of the enhanced feature vector G' is obtained as follows:

[0049]

[0050] in, is the kernel function, W' is the dimension reduction weight matrix;

[0051] The enhanced feature vector G' is input into a multi-layer fusion and convolution network to generate the fused feature vector H as follows:

[0052] where d i is the weight of the i-th mode, M and N represent the first temperature change feature and the first dynamic information feature respectively, represents a nonlinear function;

[0053] The fused feature vector H is input into the three-dimensional convolution layer, and the output value of the three-dimensional convolution layer is calculated. The expression is:

[0054] D=tanh(W*H+b), where tanh is the rectified linear activation function, W is the weight matrix of the three-dimensional convolutional layer, and b is the bias term of the three-dimensional convolutional layer.

[0055] Furthermore, the LSTM model processing process is:

[0056] S68, the LSTM model is divided into three layers, marked as the first layer, the second layer and the third layer, the first CNN data set is used to input the first layer, the first signal feature is used to input the second layer, and the first coupling feature is used to input the third layer;

[0057] S69. The final range value calculation formula of the output prediction crack of the LSTM model is:

[0058] Q=σ(W1.y LSTM1 +W2.y LSTM2 +W3.y LSTM3 +b e ),

[0059] Among them, W1, W2, and W3 represent the weight coefficients of the fully connected layers corresponding to the first, second, and third layers respectively; b e represents the bias term of the fully connected layer; y LSTM1 ,y LSTM2 ,y LSTM3 They represent the output features of the first, second and third layers respectively; σ represents the activation function, and Q represents the final range value of the output crack of the LSTM model.

[0060] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0061] The embodiment of the present invention provides a method for determining a target blasting gap range according to the geological conditions and mining requirements of a target open-pit mine; arranging a plurality of preset blasting holes in the surrounding rock, wherein the depth and spacing of the blasting holes are optimized according to the target blasting gap range; loading a preset amount of explosives in each blasting hole, and setting a plurality of segmented detonating devices, wherein the detonation time of the segmented detonating devices is controlled according to the target blasting gap range; sequentially detonating the explosives in the blasting holes by the segmented detonating devices to form a preset blasting gap range; utilizing monitoring equipment to monitor the crack expansion during the blasting process in real time, and dynamically adjusting subsequent blasting parameters according to the monitoring results; the present invention accurately predicts the blasting gap range through a neural network, and then accurately controls the blasting parameters, which has important practical significance for improving blasting efficiency, ensuring surrounding rock stability, and reducing mining costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 The present invention provides a flow chart of a method for controlling the blasting crack range of surrounding rock in an open-pit mine. DETAILED DESCRIPTION

[0063] The present invention will be described in detail below in conjunction with the accompanying drawings.

[0064] Example 1

[0065] The embodiment of the present invention provides a method for controlling the blasting gap range of surrounding rock in an open-pit mine. Figure 1 , including the following steps:

[0066] S1. Determine the target blasting gap range according to the geological conditions and mining requirements of the target open-pit mine;

[0067] S2. Arranging a plurality of preset blasting holes in the surrounding rock, wherein the depth and spacing of the blasting holes are optimized according to the target blasting gap range;

[0068] S3, loading a preset amount of explosives into each blasting hole, and setting a plurality of segmented detonating devices, wherein the detonation time of the segmented detonating devices is controlled according to the target blasting gap range;

[0069] S4, sequentially detonating the explosives in the blasting holes by means of a segmented detonation device to form a preset blasting gap range, wherein the blasting gap range is within a preset error range;

[0070] S5. Use monitoring equipment to monitor the crack expansion during the blasting process in real time, and dynamically adjust subsequent blasting parameters based on the monitoring results.

[0071] Specifically, a target blasting gap range is determined according to the geological conditions and mining requirements of the target open-pit mine; a plurality of preset blasting holes are arranged in the surrounding rock, and the depth and spacing of the blasting holes are optimized according to the target blasting gap range; a preset amount of explosives is loaded in each blasting hole, and a plurality of segmented detonating devices are provided, and the detonation time of the segmented detonating devices is controlled according to the target blasting gap range; the explosives in the blasting holes are detonated in sequence by the segmented detonating devices to form a preset blasting gap range; the crack expansion during the blasting process is monitored in real time by monitoring equipment, and subsequent blasting parameters are dynamically adjusted according to the monitoring results; the present invention accurately predicts the blasting gap range through a neural network, and then accurately controls the blasting parameters, which has important practical significance for improving blasting efficiency, ensuring surrounding rock stability and reducing mining costs.

[0072] In the above embodiment, specifically, determining the target burst gap range in S1 includes:

[0073] Determine the target blasting fracture range through numerical simulation and comprehensive analysis of data from blasting operations based on the geological conditions of the target open-pit mine, wherein the geological conditions include at least rock type, fracture development degree and joint distribution;

[0074] The specific implementation of the target burst gap range includes:

[0075] S11. Geological condition analysis: Based on the geological conditions of the target open-pit mine, generate ore body structure characteristics and rock characteristics through geological exploration;

[0076] S12, numerical simulation, using a numerical simulation method to perform N blasting numerical simulations, the input of the numerical simulation method is the geological conditions and blasting parameters of the target open-pit mine, and the output of the numerical simulation method is the simulated target blasting gap range;

[0077] S13, neural network algorithm analysis, analyzing the target blasting gap range of the target open-pit mine through the trained RBF neural network, the input of the RBF neural network is the drilling parameters, rock properties, and blasting parameters of the target open-pit mine, and the output of the RBF neural network is the RBF neural network target blasting gap range;

[0078] S14. Obtaining the target blasting gap range, calculating the first similarity between the target blasting gap range of the RBF neural network and the simulated target blasting gap range generated by N blasting numerical simulations by the cosine similarity algorithm, and presetting a similarity threshold. If the first similarity between the simulated target blasting gap range output by the nth blasting numerical simulation and the target blasting gap range of the RBF neural network meets the preset similarity threshold, then finding the simulated target blasting gap range output by the corresponding nth blasting numerical simulation as the final target blasting gap range, and at the same time finding the input blasting parameters of the corresponding nth blasting numerical simulation as the initial blasting parameters.

[0079] It should be noted that the implementation process of the cosine similarity algorithm is:

[0080]

[0081] The cosine similarity algorithm can be used to calculate the similarity of two burst gap ranges, where x i The set of feature vectors representing the burst gap range of the RBF neural network target, y i A set of characteristic vectors representing the range of the simulated target blasting gap output by the numerical simulation of the blasting sequence.

[0082] It should be noted that the specific implementation process of using the numerical simulation method to perform N-times blasting numerical simulation is as follows:

[0083] D1. Establishing numerical model:

[0084] Choose an appropriate numerical method:

[0085] Finite element method (FEM), which is suitable for dealing with mechanical problems of continuous media, can simulate the stress and strain distribution of rocks under blasting loads.

[0086] The discrete element method (DEM), suitable for simulating the behavior of granular materials, can simulate rock crushing and crack propagation.

[0087] Fluid dynamics methods (CFD), suitable for simulating the interaction between gas and rock during blasting;

[0088] This embodiment uses the finite element method to establish a numerical model;

[0089] Build the geometry model:

[0090] Using a numerical simulation method to construct a geometric model of rock and explosives, including an initial state of the rock and an arrangement of the explosives, the input of the numerical simulation method being geological conditions of a target open-pit mine and blasting parameters;

[0091] For complex fractured rock masses, the model can be analyzed and reorganized through tools such as MATLAB to construct a three-dimensional model that conforms to the actual fracture distribution law.

[0092] Define the material properties:

[0093] Define appropriate constitutive models and parameters for materials such as rock, explosives and air. For example, the HJC model can be used for rock, and the JWL equation of state can be used for explosives.

[0094] D2. Apply loads and boundary conditions:

[0095] Apply load:

[0096] Apply blasting loads, including blast stress waves and gas pressure, in numerical simulation methods;

[0097] The effect of high-pressure gas generated by explosive explosion on rock can be simulated by fluid-solid coupling method;

[0098] Define the boundary conditions:

[0099] Define the boundary conditions of the numerical simulation method according to the actual working conditions, such as free surface, fixed boundary, etc.;

[0100] D3. Numerical calculation and simulation

[0101] Choose an appropriate numerical algorithm:

[0102] For large deformation problems, the Arbitrary Lagrangian-Eulerian (ALE) algorithm can be used, which combines the advantages of the Lagrange algorithm and the Euler algorithm to avoid calculation problems caused by mesh distortion.

[0103] Perform numerical calculations:

[0104] Use finite element software (such as LS-DYNA) to perform numerical calculations to simulate stress, strain and crack propagation during blasting;

[0105] By changing the blasting parameters (such as charge amount, hole spacing, etc.), numerical simulation of blasting is performed N times;

[0106] Through numerical simulation methods, the scope and shape of blasting cracks in the surrounding rock of open-pit mines can be accurately predicted, providing theoretical support for optimizing blasting parameters and designing reasonable blasting plans. Numerical simulation methods include finite element method (FEM), discrete element method (DEM) and fluid dynamics method (CFD). Combined with fluid-solid coupling technology and appropriate numerical algorithms, they can effectively deal with complex mechanical problems in the blasting process.

[0107] The specific process of obtaining the degree of crack development and joint distribution is as follows:

[0108] S15, graying and binary image statistics are performed on the crack development degree image and joint distribution image obtained by the high-definition camera, wavelet transform series are selected, and secondary Daubechies wavelet transform is performed on the binary image statistics; one-dimensional Daubechies wavelet transform is used for processing;

[0109] S16, selecting a Gaussian filter and a filter radius to perform smoothing on the transformed image data, performing peak impact detection on the smoothed image data to obtain a peak distribution array, selecting model parameters, and using a peak impact response model to calculate an impact response of the peak distribution array;

[0110] S17, using the peak impulse response result, obtain the position when the peak distribution array response drops to 0 for the first time, obtain the most recent maximum value based on the response position, use it as the threshold, perform Daubechies wavelet inverse transform, restore it to the position of the original binary image, and use this position as the segmentation threshold of the original grayscale image to segment the image, and obtain data information corresponding to the degree of crack development and joint distribution respectively;

[0111] S18. Processing the data information corresponding to the degree of crack development and the distribution of joints through image morphology to obtain the features corresponding to the degree of crack development and the distribution of joints, respectively, wherein the features include the number of crack endpoints, the number of intersections, the surface crack rate and the fractal dimension value.

[0112] It should be noted that, in this embodiment, the data information corresponding to the degree of development of cracks and the distribution of joints is processed by the skeleton extraction algorithm in image morphology, and the image X of the data information corresponding to the degree of development of cracks and the distribution of joints is eroded and opened N times based on the hit-miss transformation, so as to finally obtain the morphological skeleton of the image of the data information corresponding to the degree of development of cracks and the distribution of joints;

[0113] The specific implementation process of step S18 is:

[0114] S181, Number of crack endpoints:

[0115] S1811, create a matrix P to store all coordinate points of the crack skeleton points, the size of the matrix P is M×2, where M represents the number of skeleton points; create an empty matrix E to record the coordinates of all skeleton endpoints;

[0116] S1812, traverse each skeleton coordinate point, if there is only one skeleton point within the eight neighborhoods of the point, record the coordinates of the point in the matrix E; if there is more than one skeleton point within the eight neighborhoods of the point, check whether there are other skeleton points in the opposite direction of each adjacent point, if not, save the coordinates of the point in the matrix E;

[0117] S1813, repeat step S1812 until all skeleton coordinate points are traversed, then all endpoint coordinates of the skeleton are saved in the matrix E, and the number of all endpoint coordinates is represented as the number of crack endpoints;

[0118] S182, number of intersection points:

[0119] Traverse each skeleton pixel point and check the number of skeleton points within its eight-neighborhood range. If there are at least three adjacent points within the eight-neighborhood range, the point satisfies the multi-connectivity feature and is a skeleton intersection point. The point is identified as a skeleton intersection point, and the total number of skeleton intersection points is expressed as the number of intersection points.

[0120] S183, surface crack rate:

[0121] Surface crack rate K t It is defined as the total length of cracks contained in the rock mass per unit area, and the formula is:

[0122]

[0123] Where A is the total area of ​​the selected rock mass image, It is expressed as the total length of the cracks in the rock mass image; for the binary rock mass crack image, the surface crack ratio is expressed by the ratio of the crack pixels to the pixels of the entire image; the rock mass crack ratio can reflect the degree of development of the rock mass structural surface. The larger the surface crack ratio, the more developed the rock mass structural surface;

[0124] S184, fractal dimension value:

[0125] This embodiment uses the box-counting fractal dimension method for calculation, and uses equal small rectangles with a side length of r to obtain the crack coverage in the image of data information corresponding to the crack development degree and joint distribution, and obtains the number N(r) of cracks passing through the small rectangles. The size of the rectangle is continuously changed to obtain the corresponding N(r). The corresponding relationship formula between N(r) and r is as follows:

[0126] N(r)=ar -D, where a is a constant, r ranges from 1 / h to 2 / h, r represents the height of the image, and D represents the fractal dimension. Finally, the corresponding relationship formula is described in the lnr-lnN coordinate system. The slope of the corresponding relationship formula is calculated using the least squares method. The absolute value of the slope represents the fractal dimension. For the rock mass structural surface, the fractal dimension value D ranges from 1.0 to 2.0. The closer the fractal dimension value D is to 1, the better the rock mass quality is. The higher the fractal dimension value D is, the worse the rock mass quality is.

[0127] In the above embodiment, specifically, the selection of the neuron parameters of the input layer, output layer and hidden layer of the RBF neural network in S13 is as follows:

[0128] Input layer: drilling speed, rotation speed, drilling impact pressure, rotation pressure, drill bit acceleration, drill bit torque, drill rod thrust, drill bit vibration state, drilling angle, drilling depth, rock type, blasting parameters, blasting effect, so the input layer has a total of 13 parameters and 13 neurons;

[0129] Output layer: predicts the target burst gap range, so the output layer has 1 parameter and 1 neuron;

[0130] Hidden layer: The number of hidden layer nodes is adjusted according to the network training results.

[0131] It should be noted that the surrounding rock blasting gap range model based on RBF neural network analyzes and studies the collected data parameter information such as drilling speed, rotation speed, drilling impact pressure, rotation pressure, drill bit acceleration, drill bit torque, drill rod thrust, drill bit vibration state, drilling angle, drilling depth, rock type, blasting parameters, blasting effect, etc., establishes a neural network model, and outputs the surrounding rock blasting gap range information.

[0132] Since the rock blasting process is highly complex and nonlinear, neural networks can play a huge role in the modeling process. Since neural networks can directly build models based on the input / output data of the object, they do not require knowledge of the object and complex mathematical formula derivation, and the network learning accuracy target can be achieved by using appropriate training algorithms. Therefore, it is very effective and relatively easy to implement system modeling using neural network models. A neural network model of drilling speed, rotation speed, drilling impact pressure, rotation pressure, drill bit acceleration, drill bit torque, drill rod thrust, drill bit vibration state, drilling angle, drilling depth, rock type, blasting parameters, blasting effect and surrounding rock blasting gap range was constructed.

[0133] A three-layer neural network is used to construct a prediction and design model for the range of surrounding rock bursting gaps (input layer, hidden layer and output layer). The process is to first train and learn the network through a large number of engineering example training samples, select the appropriate number of hidden layer nodes, training accuracy, and number of iterations, so as to obtain the optimal transfer coefficient of the network node function, and then perform inference calculation based on the input parameters to obtain the optimal output.

[0134] A hyperbolic tangent function is used between the input layer and the hidden layer, a linear rectification function ReLU is used between the hidden layer and the output layer, and a gradient descent function Adam is selected as a training function.

[0135] The value range of the number of neurons in the hidden layer is calculated as follows:

[0136]

[0137] In the formula, H represents the number of neurons in the hidden layer, a represents the number of neurons in the input layer, b represents the number of neurons in the output layer, and c is a positive integer between 1 and 10.

[0138] A neural network model is established, and a three-layer neural network is used to construct the design model (input layer, hidden layer and output layer). Its input parameters are: relevant parameters obtained during drilling and relevant parameters before and after blasting, and the output result is: information on the range of blasting gaps in the surrounding rock. This neural network model is composed of an RBF neural network structure. The process is to collect a large amount of actual engineering data to train and learn the network, so as to obtain the optimal transfer coefficient of the network node function, and then perform reasoning calculation based on the input parameters (relevant parameters obtained during drilling and relevant parameters before and after blasting, a total of 13 parameters, that is, n = 13) to obtain the optimal output.

[0139] In the above embodiment, specifically, using monitoring equipment to monitor the crack expansion during the blasting process in real time in S5 includes:

[0140] S54, using a distributed optical fiber sensor to monitor the temperature change inside the surrounding rock, inverting the expansion of the crack through the temperature change, and generating a first temperature change feature;

[0141] S55. Use optical fiber sensors and strain gauges to monitor strain changes on the surface of the surrounding rock, obtain dynamic information on crack expansion in real time, and generate a first dynamic information feature.

[0142] In the above embodiment, specifically, using monitoring equipment to monitor the crack expansion in real time during the blasting process in S5 also includes:

[0143] S56, using an acoustic emission monitoring sensor to monitor the acoustic emission signal of the surrounding rock in real time during the blasting process, determining the expansion position and rate of the crack through signal analysis, and generating a first signal feature;

[0144] S57. Use a multi-physical field coupling monitoring sensor to simultaneously monitor the stress, strain, temperature and acoustic emission signals of the surrounding rock to generate a first coupling feature.

[0145] It should be noted that the first temperature change feature, the first dynamic information feature, the first signal feature and the first coupling feature are all extracted in the following manner:

[0146] L1. Preprocess and mark the first temperature change feature, the first dynamic information feature, the first signal feature and the first coupling feature before feature extraction, use statistical methods to detect and remove abnormal values, normalize the data of different sensors, synchronize all data according to timestamps, and mark them as first temperature change data, first dynamic information data, first signal data and first coupling data respectively;

[0147] L2. Draw waveform signal curve graphs for the first temperature change data, the first dynamic information data, the first signal data, and the first coupling data, respectively, calculate the signal frequency fluctuation information of the first temperature change data, the first dynamic information data, the first signal data, and the first coupling data according to the waveform signal curve graph, determine the abnormal signal position according to the signal frequency fluctuation information, perform waveform extraction on the abnormal signal position in the waveform signal curve graph, and determine the noise type existing in the first temperature change data, the first dynamic information data, the first signal data, and the first coupling data;

[0148] L3. Determine a signal noise removal method for the first temperature change data, the first dynamic information data, the first signal data, and the first coupling data according to the existing noise type, and perform exception processing on the first temperature change data, the first dynamic information data, the first signal data, and the first coupling data according to the signal noise removal method;

[0149] L4, dividing the first temperature change data, the first dynamic information data, the first signal data and the first coupling data after the abnormality processing into N equal parts of waveform signal data blocks, converting N into binary representation, and determining the binary bit length of N;

[0150] L5, performing CWT continuous wavelet transform operation on the waveform signal data block based on wavelet transform and wavelet function, and then performing scaling and translation to obtain frequency domain data of first temperature change data, first dynamic information data, first signal data and first coupling data respectively;

[0151] L6. Plotting the frequency domain data of the first temperature change data, the first dynamic information data, the first signal data and the first coupling data into a waveform signal spectrum diagram;

[0152] L7. Calculate the peak frequency, frequency bandwidth, center frequency, and spectrum energy of the waveform signal spectrum diagram to obtain spectrum characteristics of the waveform signal spectrum diagram. The spectrum characteristics of the waveform signal spectrum diagram are respectively represented by a first temperature change characteristic, a first dynamic information characteristic, a first signal characteristic, and a first coupling characteristic.

[0153] Specifically, the extracted first temperature change feature, the first dynamic information feature, the first signal feature and the first coupling feature are divided into a training set and a test set, and the training set data is used to train the first deep learning algorithm. After the training is completed, the feature importance evaluation method provided by the first deep learning algorithm is used to identify the features that have the greatest impact on the final range value of the crack; the test set is input into the trained first deep learning algorithm, the prediction result of each sample is obtained, the final range value of the crack is predicted, and an output list of the final range value of the crack is formed, and the performance of the model is evaluated using accuracy, recall and F1-score. In this embodiment, the extracted features are divided into a training set (70%) and a test set (30%).

[0154] In the above embodiment, specifically, dynamically adjusting subsequent blasting parameters according to the monitoring results in S5 includes:

[0155] S61. According to the real-time monitoring of the crack expansion, the amount of explosives, detonation time and hole spacing of subsequent blasting holes are adjusted to ensure that the crack range meets the target requirements;

[0156] S62, by establishing a mathematical model between blasting parameters and crack range, using a first blasting parameter analysis algorithm to analyze the first temperature change feature, the first dynamic information feature, the first signal feature and the first coupling feature, and predicting the optimal blasting parameters;

[0157] S63. After each blasting, the blasting effect is evaluated, and the parameters of subsequent blasting are adjusted according to the evaluation results to form a closed-loop control.

[0158] Specifically, the specific implementation process of the first blasting parameter analysis algorithm is:

[0159] In the process of predicting the optimal blasting parameter combination, the blasting parameters and the corresponding first temperature change characteristics, first dynamic information characteristics, first signal characteristics and first coupling characteristics are integrated into a data set to ensure that each set of blasting parameters corresponds to the subsequent crack range; the Pearson correlation coefficient is used to calculate the correlation between each blasting parameter and the open-pit mine surrounding rock blasting monitoring parameter, generate a correlation matrix, and identify positive and negative correlations. In this embodiment, the correlation coefficient between the unit consumption of explosives and the stress of the open-pit mine surrounding rock wall is 0.86 (strong positive correlation), and the correlation coefficient between the detonation interval time and the vibration acceleration is -0.78 (negative correlation); the association rule learning algorithm Apriori is used to mine the relationship between the blasting parameters and the open-pit mine surrounding rock blasting monitoring parameters, and identify potential association rules; through association rule learning, it is found that when the unit consumption of explosives is greater than 0.7kg / m and the peripheral eye spacing is less than 0.5m, the risk of over-limit stress of the open-pit mine surrounding rock wall increases significantly (confidence is 80%);

[0160] Bayesian linear regression was selected for modeling. The open-pit mine surrounding rock blasting monitoring parameters were used as dependent variables and blasting parameters as independent variables. The data was divided into training set and test set. The Bayesian linear regression model was trained using the training set and verified on the test set to evaluate the model's fitting degree and prediction ability. The coefficients of the Bayesian linear regression model were used to quantitatively evaluate the influence of each blasting parameter on the open-pit mine surrounding rock blasting monitoring parameters. The larger the absolute value of the coefficient, the greater the influence of the blasting parameter on the open-pit mine surrounding rock blasting monitoring parameters.

[0161] The decision tree model has good performance and can accurately predict the blasting monitoring parameters of the open-pit mine surrounding rock after blasting. By combining the output of the decision tree model with expert advice, the optimal blasting parameter scheme under different parameter combinations was identified. The results of linear regression analysis show that the unit consumption of explosives, the detonation interval, and the peripheral eye spacing have a significant impact on the monitoring parameters of the open-pit mine surrounding rock blasting. The unit consumption of explosives has the greatest positive impact on the stress of the open-pit mine surrounding rock wall, and the detonation interval has a significant negative impact on the vibration acceleration. The association rule mining results provide multiple potential parameter combination rules. For example, a combination of higher unit consumption of explosives and smaller peripheral eye spacing may increase the risk of excessive stress on the open-pit mine surrounding rock wall.

[0162] In the above embodiment, specifically, in S63, after each blasting, the blasting effect is evaluated, and the parameters of the subsequent blasting are adjusted according to the evaluation result, so as to form a closed-loop control including:

[0163] S64, using a drone equipped with a high-resolution camera to quickly scan the surface of the surrounding rock after blasting, obtain three-dimensional distribution information of cracks, and generate a first three-dimensional image feature;

[0164] S65, using a first deep learning algorithm to perform real-time analysis on the first temperature change feature, the first dynamic information feature, the first signal feature, the first coupling feature, and the first three-dimensional image feature, identify key features of crack extension, and predict a final range of the crack; the first temperature change feature, the first dynamic information feature, the first signal feature, the first coupling feature, and the first three-dimensional image feature are divided into a training set and a test set, and the training set and the test set are used for training the first deep learning algorithm;

[0165] S66. Automatically adjust the parameters of subsequent blasting according to the prediction results.

[0166] It should be noted that, according to the prediction results, the parameters of subsequent blasting are automatically adjusted, and the parameters are automatically sent to the on-site staff through the system for optimization suggestions, and the on-site staff adjust the blasting parameters according to the suggestions.

[0167] It should be noted that, in this embodiment, some blasting parameters are as follows: charge: 0.5-1.5 kg / m, peripheral hole spacing: 0.3-0.8 m, number of blast holes: 2-6, blast hole depth: 2-3.5 m, explosive unit consumption: 0.4-1.0 kg / m 3 , detonation interval: 0.001~0.005s. Initial safety index (SI): open-pit mine surrounding rock wall stress: 11.2MPa, open-pit mine surrounding rock displacement: 2.8mm, vibration acceleration: 30m / s2, crack area range: 14.41~16.73 times the blast hole radius.

[0168] Parameters before optimization: Charge: 1.2kg / m, Peripheral hole spacing: 0.5m, Number of blastholes: 4, Blast hole depth: 3.0m, Explosive consumption per unit: 0.85kg / m 3 , detonation interval time: 0.003s, crack zone range: 15.83 times the blast hole radius.

[0169] The monitoring results are as follows: open-pit mine surrounding rock wall stress: 11.2MPa, open-pit mine surrounding rock displacement: 2.8mm, vibration acceleration: 30m / s2.

[0170] Parameters after the first deep learning algorithm optimization: Charge: 1.0kg / m, Peripheral hole spacing: 0.45m, Number of blastholes: 2, Explosive consumption per unit: 0.78kg / m 3 , detonation interval: 0.0025s, crack zone range: 14.56 times the blast hole radius.

[0171] The monitoring results are as follows: open-pit mine surrounding rock wall stress: 9.8MPa (reduced by 12.5%), open-pit mine surrounding rock displacement: 2.1mm (reduced by 25%), vibration acceleration: 25m / s2 (reduced by 16.7%), fracture zone range: 14.63 times the blasthole radius (reduced by 11.7%).

[0172] In the above embodiment, specifically, the first deep learning algorithm includes:

[0173] The first deep learning algorithm uses a convolutional neural network (CNN) combined with a long short-term memory (LSTM) network; the training set is analyzed by a CNN-LSTM model to obtain a final model after the CNN-LSTM model is trained, and the convolutional neural network (CNN) uses a DenseNet neural network framework; the long short-term memory (LSTM) network is used to capture the dependency relationship between the accumulation of the first temperature change feature, the first dynamic information feature, the first signal feature, the first coupling feature, and the first three-dimensional image feature and time, and then output the final range value of the crack;

[0174] The specific implementation process of the CNN-LSTM model is as follows:

[0175] The CNN-LSTM model architecture and processing process are as follows: the CNN model and the LSTM model are connected in series, the first three-dimensional image feature is input into the CNN model, the first CNN data set is output through CNN model feature processing, and the first temperature change feature, the first dynamic information feature, the first signal feature, and the first coupling feature are input into the LSTM model;

[0176] The processing process of the CNN model is:

[0177] Applying kernel K-means clustering analysis with adaptive weight allocation to the first three-dimensional image feature Y, feature enhancement and dimensionality reduction are performed, and the expression of the enhanced feature vector G' is obtained as follows:

[0178]

[0179] in, is the kernel function, W' is the dimension reduction weight matrix;

[0180] The enhanced feature vector G' is input into a multi-layer fusion and convolution network to generate the fused feature vector H as follows:

[0181] where d i is the weight of the i-th mode, M and N represent the first temperature change feature and the first dynamic information feature respectively, represents a nonlinear function;

[0182] The fused feature vector H is input into the three-dimensional convolution layer, and the output value of the three-dimensional convolution layer is calculated. The expression is:

[0183] D=tanh(W*H+b), where tanh is the rectified linear activation function, W is the weight matrix of the three-dimensional convolutional layer, and b is the bias term of the three-dimensional convolutional layer.

[0184] In the above embodiment, specifically, the LSTM model processing process is:

[0185] S68, the LSTM model is divided into three layers, marked as the first layer, the second layer and the third layer, the first CNN data set is used to input the first layer, the first signal feature is used to input the second layer, and the first coupling feature is used to input the third layer;

[0186] S69. The final range value calculation formula of the output prediction crack of the LSTM model is:

[0187] Q=σ(W1.y LSTM1 +W2.y LSTM2 +W3.y LSTM3 +b e ),

[0188] Among them, W1, W2, and W3 represent the weight coefficients of the fully connected layers corresponding to the first, second, and third layers respectively; b e represents the bias term of the fully connected layer; y LSTM1 ,y LSTM2 ,y LSTM3 They represent the output features of the first, second and third layers respectively; σ represents the activation function, and Q represents the final range value of the output crack of the LSTM model.

[0189] It should be noted that adaptive weights are introduced for W1, W2, and W3 to satisfy the following formula:

[0190]

[0191] in, represents the adaptive weight of the i-th layer at the t-th iteration, T represents the maximum number of iterations, and W max Represents the maximum weight coefficient, W min represents the minimum weight coefficient, represents the maximum adaptive inertia value at the tth iteration, represents the minimum adaptive inertia value at the tth iteration.

[0192] It should be understood that the above-mentioned embodiments are one or more embodiments of the present invention, and there are many other embodiments and variations thereof based on the present invention; the variations and modifications made by ordinary technicians in this industry through the present invention without making pioneering innovations all fall within the scope of protection of the present invention.

Claims

1. A method for controlling the blasting crack range of surrounding rock in an open-pit mine, characterized in that: The following steps are involved: S1. Determine the target blasting gap range and initial blasting parameters according to the geological conditions and mining requirements of the target open-pit mine; S2. Arranging a plurality of preset blasting holes in the surrounding rock, wherein the depth and spacing of the blasting holes are designed according to the initial blasting parameters; S3, loading a preset amount of explosives into each blasting hole, and setting a plurality of segmented detonating devices, wherein the detonation time of the segmented detonating devices is controlled according to the initial blasting parameters; S4, sequentially detonating the explosives in the blasting holes by means of a segmented detonation device to form a preset blasting gap range, wherein the blasting gap range is within a preset error range; S5. Use monitoring equipment to monitor the crack expansion during the blasting process in real time, and dynamically adjust subsequent blasting parameters based on the monitoring results.

2. A method for controlling the blasting crack range of surrounding rock in an open-pit mine according to claim 1, characterized in that: The target blasting gap range determined in S1 includes: Determine the target blasting fracture range through numerical simulation and comprehensive analysis of data from blasting operations based on the geological conditions of the target open-pit mine, wherein the geological conditions include at least rock type, fracture development degree and joint distribution; The specific implementation of the target burst gap range includes: S11. Geological condition analysis: Based on the geological conditions of the target open-pit mine, generate ore body structure characteristics and rock characteristics through geological exploration; S12, numerical simulation, using a numerical simulation method to perform N blasting numerical simulations, the input of the numerical simulation method is the geological conditions and blasting parameters of the target open-pit mine, and the output of the numerical simulation method is the simulated target blasting gap range; S13, neural network algorithm analysis, analyzing the target blasting gap range of the target open-pit mine through the trained RBF neural network, the input of the RBF neural network is the drilling parameters, rock properties, and blasting parameters of the target open-pit mine, and the output of the RBF neural network is the RBF neural network target blasting gap range; S14. Obtaining the target blasting gap range, calculating the first similarity between the target blasting gap range of the RBF neural network and the simulated target blasting gap range generated by N blasting numerical simulations by the cosine similarity algorithm, and presetting a similarity threshold. If the first similarity between the simulated target blasting gap range output by the nth blasting numerical simulation and the target blasting gap range of the RBF neural network meets the preset similarity threshold, then finding the simulated target blasting gap range output by the corresponding nth blasting numerical simulation as the final target blasting gap range, and at the same time finding the input blasting parameters of the corresponding nth blasting numerical simulation as the initial blasting parameters.

3. A method for controlling the blasting crack range of surrounding rock in an open-pit mine according to claim 2, characterized in that: The specific process of obtaining the degree of fissure development and joint distribution is as follows: S15, graying and binary image statistics are performed on the crack development degree image and joint distribution image obtained by the high-definition camera, wavelet transform series are selected, and secondary Daubechies wavelet transform is performed on the binary image statistics; one-dimensional Daubechies wavelet transform is used for processing; S16, selecting a Gaussian filter and a filter radius to perform smoothing on the transformed image data, performing peak impact detection on the smoothed image data to obtain a peak distribution array, selecting model parameters, and using a peak impact response model to calculate an impact response of the peak distribution array; S17, using the peak impulse response result, obtain the position when the peak distribution array response drops to 0 for the first time, obtain the most recent maximum value according to the response position, use it as the threshold, perform Daubechies wavelet inverse transform, restore it to the position of the original binary image, and use this position as the segmentation threshold of the original grayscale image to segment the image, and obtain data information corresponding to the degree of crack development and joint distribution respectively; S18. Processing the data information corresponding to the degree of crack development and the distribution of joints through image morphology to obtain the features corresponding to the degree of crack development and the distribution of joints, respectively, wherein the features include the number of crack endpoints, the number of intersections, the surface crack rate and the fractal dimension value.

4. A method for controlling the blasting crack range of surrounding rock in an open-pit mine according to claim 2, characterized in that: The selection of neuron parameters of the input layer, output layer and hidden layer of the RBF neural network in S13 is as follows: Input layer: drilling speed, rotation speed, drilling impact pressure, rotation pressure, drill bit acceleration, drill bit torque, drill rod thrust, drill bit vibration state, drilling angle, drilling depth, rock type, blasting parameters, blasting effect, so the input layer has a total of 13 parameters and 13 neurons; Output layer: predicts the target burst gap range, so the output layer has 1 parameter and 1 neuron; Hidden layer: The number of hidden layer nodes is adjusted according to the network training results.

5. The method for controlling the blasting crack range of surrounding rock in an open-pit mine according to claim 1, characterized in that: In S5, the monitoring equipment is used to monitor the crack expansion during the blasting process in real time, including: S54, using a distributed optical fiber sensor to monitor the temperature change inside the surrounding rock, inverting the expansion of the crack through the temperature change, and generating a first temperature change feature; S55. Use optical fiber sensors and strain gauges to monitor strain changes on the surface of the surrounding rock, obtain dynamic information on crack expansion in real time, and generate a first dynamic information feature.

6. A method for controlling the blasting crack range of surrounding rock in an open-pit mine according to claim 5, characterized in that: In S5, the use of monitoring equipment to monitor the crack expansion during the blasting process in real time also includes: S56, using an acoustic emission monitoring sensor to monitor the acoustic emission signal of the surrounding rock in real time during the blasting process, determining the expansion position and rate of the crack through signal analysis, and generating a first signal feature; S57. Use a multi-physical field coupling monitoring sensor to simultaneously monitor the stress, strain, temperature and acoustic emission signals of the surrounding rock to generate a first coupling feature.

7. A method for controlling the blasting crack range of surrounding rock in an open-pit mine according to claim 6, characterized in that: In S5, the subsequent blasting parameters are dynamically adjusted according to the monitoring results, including: S61. According to the real-time monitoring of the crack expansion, the amount of explosives, the detonation time and the hole spacing of the subsequent blasting holes are adjusted to ensure that the crack range meets the target requirements; S62, by establishing a mathematical model between blasting parameters and crack range, using a first blasting parameter analysis algorithm to analyze the first temperature change feature, the first dynamic information feature, the first signal feature and the first coupling feature, and predicting the optimal blasting parameters; S63. After each blasting, the blasting effect is evaluated, and the parameters of subsequent blasting are adjusted according to the evaluation results to form a closed-loop control.

8. A method for controlling the blasting crack range of surrounding rock in an open-pit mine according to claim 7, characterized in that: In S63, after each blasting, the blasting effect is evaluated, and the parameters of the subsequent blasting are adjusted according to the evaluation results, forming a closed-loop control including: S64, using a drone equipped with a high-resolution camera to quickly scan the surface of the surrounding rock after blasting, obtain three-dimensional distribution information of cracks, and generate a first three-dimensional image feature; S65, using a first deep learning algorithm to perform real-time analysis on the first temperature change feature, the first dynamic information feature, the first signal feature, the first coupling feature, and the first three-dimensional image feature, identify key features of crack extension, and predict a final range of the crack; the first temperature change feature, the first dynamic information feature, the first signal feature, the first coupling feature, and the first three-dimensional image feature are divided into a training set and a test set, and the training set and the test set are used for training the first deep learning algorithm; S66. Automatically adjust the parameters of subsequent blasting according to the prediction results.

9. A method for controlling the blasting crack range of surrounding rock in an open-pit mine according to claim 8, characterized in that: The first deep learning algorithm includes: The first deep learning algorithm uses a convolutional neural network (CNN) combined with a long short-term memory (LSTM) network; the training set is analyzed by a CNN-LSTM model to obtain a final model after the CNN-LSTM model is trained, and the convolutional neural network (CNN) uses a DenseNet neural network framework; the long short-term memory (LSTM) network is used to capture the dependency relationship between the accumulation of the first temperature change feature, the first dynamic information feature, the first signal feature, the first coupling feature, and the first three-dimensional image feature and time, and then output the final range value of the crack; The specific implementation process of the CNN-LSTM model is as follows: The CNN-LSTM model architecture and processing process are as follows: the CNN model and the LSTM model are connected in series, the first three-dimensional image feature is input into the CNN model, the first CNN data set is output through CNN model feature processing, and the first temperature change feature, the first dynamic information feature, the first signal feature, and the first coupling feature are input into the LSTM model; The processing process of the CNN model is: Applying kernel K-means clustering analysis with adaptive weight allocation to the first three-dimensional image feature Y, feature enhancement and dimensionality reduction are performed, and the expression of the enhanced feature vector G' is obtained as follows: in, is the kernel function, W' is the dimension reduction weight matrix; The enhanced feature vector G' is input into a multi-layer fusion and convolution network to generate the fused feature vector H as follows: where d i is the weight of the i-th mode, M and N represent the first temperature change feature and the first dynamic information feature respectively, represents a nonlinear function; The fused feature vector H is input into the three-dimensional convolution layer, and the output value of the three-dimensional convolution layer is calculated. The expression is: D=tanh(W*H+b), where tanh is the rectified linear activation function, W is the weight matrix of the three-dimensional convolutional layer, and b is the bias term of the three-dimensional convolutional layer.

10. A method for controlling the blasting crack range of surrounding rock in an open-pit mine according to claim 9, characterized in that: The LSTM model processing process is: S68, the LSTM model is divided into three layers, marked as the first layer, the second layer and the third layer, the first CNN data set is used to input the first layer, the first signal feature is used to input the second layer, and the first coupling feature is used to input the third layer; S69. The final range value calculation formula of the output prediction crack of the LSTM model is: Q=σ(W1.y LSTM1 +W2.y LSTM2 +W3.y LSTM3 +b e ), Among them, W1, W2, and W3 represent the weight coefficients of the fully connected layers corresponding to the first, second, and third layers respectively; b e represents the bias term of the fully connected layer; y LSTM1 ,y LSTM2 ,y LSTM3 They represent the output features of the first, second and third layers respectively; σ represents the activation function, and Q represents the final range value of the output crack of the LSTM model.

Images