Blasting vibration prediction method for optimizing BP neural network based on atomic search algorithm
By introducing atomic search algorithm (ASO) to optimize network structure and parameters, the problem of low learning efficiency and easy to fall into local extreme values in blast vibration prediction is solved, and more efficient and accurate prediction effects are achieved.
Patent Information
- Application Number
- CN202510076148.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-17
- Publication Date
- 2025-05-06
AI Technical Summary
The existing blasting vibration prediction method based on BP neural network has problems such as low learning efficiency, slow convergence speed and easy to fall into local extreme values.
Atomic search algorithm (ASO) is used to optimize the BP neural network, and by building an ASO-BP burst vibration speed prediction model, the network topology, weights and thresholds are optimized, and the generalization ability and prediction accuracy of the model are improved.
It improves the generalization ability and prediction accuracy of the model, can automatically learn and identify the blasting vibration speed law, and is suitable for nonlinear and time-varying complex sample parameters.
Smart Images
Figure CN119940129A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of blasting vibration safety technology, and in particular to a blasting vibration prediction method based on an atomic search algorithm to optimize a BP neural network. Background Art
[0002] The scale of open-pit coal mining is expanding day by day, and the scale of single blasting is mostly above one million cubic meters. The vibration effect caused by blasting has an increasingly obvious impact on the surrounding environment. At present, the formula empirical method is still used to predict the peak velocity (PPV) of blasting vibration. Based on a large number of engineering practices, several factors affecting blasting vibration are summarized. Through dimensional analysis and combined with the correction of on-site conditions, the empirical calculation formula of blasting vibration intensity is obtained, such as the national standard "Safety Regulations for Blasting" GB6722-2014. The prediction of blasting vibration in the middle and far areas can still meet the needs of engineering projects. However, with the expansion of the scale of mining and stripping, the application of new technologies such as high slope and steep wall mining, the requirements for mine slope management are becoming more and more stringent, and the prediction error of blasting vibration in the near area makes the above empirical formula completely inapplicable.
[0003] Based on this, it is particularly important to study a method suitable for predicting blasting vibration in large-scale mines. At present, domestic and foreign researchers have proposed many new methods for predicting blasting vibration. In particular, the rapid development of fields such as artificial intelligence and machine learning has provided us with neural networks, support vector machine algorithms, swarm intelligence, and genetic algorithms. Among them, BP neural network is the most commonly used method for blasting vibration prediction, but considering that BP neural network has problems such as low learning efficiency, slow convergence speed, and easy to fall into local extreme values. Summary of the invention
[0004] The present application provides a blasting vibration prediction method based on an atomic search algorithm to optimize a BP neural network, so as to solve the technical problems of low learning efficiency, slow convergence speed and easy falling into local extreme values in the existing blasting vibration prediction method based on a BP neural network.
[0005] The present application provides a method for predicting blasting vibration based on an atomic search algorithm to optimize a BP neural network, comprising the following steps: Establishing a blasting vibration sample database, wherein the input sample data of the blasting vibration sample database includes blasting center distance, elevation difference, hole spacing, row spacing, hole depth, single hole charge, maximum section charge, delay time, minimum resistance line, and rock density; Obtaining output result sample data, wherein the output result sample data includes the peak vibration velocity of the particle at the measured point in three directions of XYZ; Preprocessing the input sample data and the output result sample data; Construct a BP neural network and obtain the number of hidden layers and neurons in the hidden layer; Based on the ASO algorithm, the BP neural network is optimized, the ASO-BP blasting vibration velocity prediction model is constructed, and the network topology is determined; Input the acquisition parameters into the ASO-BP blasting vibration velocity prediction model and obtain the blasting vibration velocity prediction data.
[0006] Optionally, the step of preprocessing the input sample data and the output result sample data comprises the steps of: The step of preprocessing the input sample data and the output result sample data comprises the steps of: The input sample data and the output result sample data are normalized: Among them, in the formula: is the normalized value. The normalized data are all in the interval \left [ {-1,1} \right ] Inside, is the value of the factor to be normalized, is the minimum value of this factor in the sample database, is the maximum value of this factor in the sample database.
[0007] Optionally, the step of constructing a BP neural network and obtaining the number of hidden layers and the number of neurons in the hidden layer specifically includes the following steps: According to Kolmogorov theorem, a three-layer neural network model is constructed using a single hidden layer, and the network structure of the three-layer neural network model includes an input layer, an output layer and a hidden layer; Determine the number of hidden layer nodes: in, is the number of nodes in the input layer; is the number of hidden layer nodes; is the number of nodes in the output layer; It is the adjustment constant between hidden layers and takes an integer between 1 and 10.
[0008] Optionally, the step of constructing a three-layer neural network model using a single hidden layer according to the Kolmogorov theorem includes: According to Kolmogorov's theorem, given any continuous function f:{U}^{n}\to {R}^{m},f\left ( {x} \right )=y,U\in \left [ {0,1} \right ] , Indicates from A subset of 2-dimensional Euclidean space arrive dimensional Euclidean space subset The mapping of is the unit closed interval \left [ {0,1} \right ] , continuous function It is implemented using a 3-layer feedforward neural network. The input layer of this network has neurons, and the hidden layer has neurons, and the third layer (output layer) has A neuron.
[0009] Optionally, the step of optimizing the BP neural network based on the ASO algorithm specifically includes the following steps: Using the pre-processed input sample data and the output result sample data to perform initial network training on the BP neural network to obtain network weights and thresholds; The network weights and thresholds are encoded with real values to form initial individuals and obtain the initial atomic population; The training error of the BP neural network is used as the individual fitness value, that is, the mean absolute error of the training set and the test set is selected as the fitness of ASO, and the fitness function is obtained: , in, is the expected output value of the BP neural network, is the actual output value of the BP neural network, - represents the training error of BP neural network, represents the sample size of the training set, express Function to get the dimension of an array. Represents the fitness value, which is the standard for evaluating the quality of individuals; The ASO algorithm is used to update the atomic population and the fitness function is used to calculate the fitness value of individuals in the updated atomic population; By iteratively calculating the best fitness of each atom, selecting the best individual, and optimizing the BP neural network, Dai Zaiwei Previous The acceleration of an atom is: in, Indicates The acceleration of atoms, Indicates The interaction force on each atom is Indicates The covalent bond forces on atoms Indicates The mass of an atom; No. In the iteration The mass of an atom It is determined by the fitness value of the individual atomic group, and its definition is as follows: , Where: Indicates The estimated mass of atoms, and Respectively indicate the The objective function fitness values of the best and worst atoms during iteration; Indicates Atom in The function fitness value at the iteration; represents the total number of atoms, Indicates atoms, To simplify the ASO algorithm, in the global optimization process of the algorithm, at the t+1th iteration i The position and velocity of an atom in the dth dimension are expressed as: , in, Indicates Generation The first The speed of atoms, Represents the interval \left [ {0,1} \right ] A random number between Indicates Generation Wei Shangdi The speed of atoms, Indicates Generation Wei Shangdi The acceleration of atoms, Indicates Generation Wei Shangdi The position of atoms, Indicates Generation Wei Shangdi The position of atoms; The BP neural network is trained and tested using weights and thresholds optimized by the APO algorithm.
[0010] Optionally, by iteratively calculating the best fitness of each atom, selecting the best individual, and optimizing the BP neural network, Get the The interaction forces on atoms The solution formula is: , Where: interaction force Indicates the surrounding atoms The sum of the forces acting on the atoms is represents the current search space dimension, ; Indicates the current iteration number; is a subset of the total number of atoms, indicating The set of atoms with the best fitness function value; is the interval \left [ {0,1} \right ] A random number between Indicates In the iteration, Atomic pair Lennard-Jones potential force of atoms; The Lennard-Jones potential force is expressed as: ,, Where: interaction force Indicates the surrounding atoms The sum of the forces acting on the atoms is is a depth function used to adjust the repulsive and attractive areas. For the Atoms and The scaled distance between atoms, Indicates the current iteration number; is the depth function, as shown below: Where: is the depth weight, is the maximum number of iterations, Indicates the current iteration number; For the Atoms and The scaled distance between atoms is set to , the upper limit of attraction with a larger function value is set to , which is defined as: , Where: and They are The lower and upper limits of and The definition of is as follows: , Where: , Respectively represent the Atoms and The position of atoms; is the Euclidean distance between atoms; Indicates the current iteration number; , Where: , Respectively represent the Atoms and The position of atoms; represents the collision scale, is a subset of the total number of atoms, indicating The set of atoms with the best fitness function value, Indicates the current iteration number; is a time function, which decreases as the number of iterations increases. The definition of is as follows: Where: is the total number of atoms; is the total number of iterations, Indicates the current iteration number.
[0011] Optionally, by iteratively calculating the best fitness of each atom, selecting the best individual, and optimizing the BP neural network, Assuming that each atom has a covalent bond with the best atom and each atom is bound by the best atom, The atoms are constrained to , Where: It is The position of the best atom at iteration ; It is The square of the fixed bond length between the atom and the optimal atom, then The binding force of atoms is , Where: Represents the current search space dimension, Indicates the number of times in the current search space dimension Iteration Atomic constraints, is the dimension of the current search space. The best atomic position of the iteration, is the dimension of the current search space. Atom The position of the iteration, is the Lagrange multiplier, , is the multiplier weight, is the total number of iterations; let , then the constraint force caused by the covalent bond is solved as follows: , Where: is the multiplier weight, is the best atomic position for the tth iteration in the current search space dimension, is the dimension of the current search space. Atom The position of the iteration, is the total number of iterations, so under the interaction force and geometric constraints, Atoms in The acceleration at this moment is: , in, Indicates The acceleration of atoms, Indicates The interaction force on each atom is Indicates The covalent bond forces on atoms Indicates The mass of an atom, is the depth weight, Indicates the current iteration number, is the total number of iterations, represents the current search space dimension, , is a subset of the total number of atoms, indicating The set of atoms with the best fitness function value; is the interval \left [ {0,1} \right ] A random number between Indicates In the iteration, Atomic pair The Lennard-Jones potential force of atoms, is the dimension of the current search space. Atom The position of the iteration, is the dimension of the current search space. Atom The position of the iteration, For the The best atomic position of the iteration, is the multiplier weight.
[0012] Optionally, using the pre-processed input sample data and the output result sample data to perform initial network training on the BP neural network to obtain network weights and thresholds includes: In the interval \left [ {-1,1} \right ] Initialize the network weights randomly , , threshold 、 , Represents the input layer neurons and hidden layer The connection weights between neurons; represents the hidden layer The neurons in the output layer The connection weights between neurons; Represents the threshold of the hidden layer neurons ， represents the threshold of the output layer neuron; Indicates the number of input layer nodes; Represents the number of hidden layer nodes; , is the number of nodes in the output layer; Assume that all hidden layer neurons use activation function function: , in, Represents the input signal of the hidden layer; Calculate the output value of the hidden layer neuron ; , in, represents the hidden layer activation function; Indicates that the neuron receives The input signal passed from the input layer; Represents the input layer neurons and hidden layer The connection weights between neurons; Indicates Input signal, represents the hidden layer The threshold of each neuron; ， is the number of hidden layer nodes; Calculate the output value of the output layer neuron ; , in, represents the output value of the output layer neuron, represents the output layer activation function, ， is the number of hidden layer nodes, ， is the number of nodes in the output layer; represents the hidden layer neurons and the output layer The connection weights between neurons; represents the output value of the hidden layer neuron, Represents the output layer The threshold of each neuron; The error function is defined as: , in, Indicates the number of training samples; is the number of training sample groups; ， is the number of nodes in the output layer, represents the predicted value of the output layer, Indicates actual value; Update the connection weight to: , in, is the learning rate parameter, in the interval \left [ {0,1} \right ] Take values between Indicates At the iteration, the hidden layer The neurons in the output layer The connection weights between neurons; represents the output value of the hidden layer neuron, , is the number of nodes in the output layer, Indicates Input signal, represents the output value of the output layer neuron, represents the predicted value of the output layer, Indicates the number of iterations; Update thresholds: , in, Indicates At the iteration, the hidden layer The threshold of each neuron; The output layer The threshold of the neuron; is the learning rate parameter, in the interval \left [ {0,1} \right ] Take values between represents the output value of the hidden layer neuron, Indicates Input signal, represents the hidden layer The neurons in the output layer The connection weights between neurons; represents the output value of the output layer neuron, represents the predicted value of the output layer neuron, Indicates the number of iterations; Determine whether the iteration is finished. If not, return to the step of calculating the output value of the hidden layer neuron.
[0013] Optionally, in the step of determining whether the iteration is ended, a condition for ending the iteration is that a preset maximum number of iterations is reached or a fitness value reaches a preset threshold.
[0014] Correspondingly, the present application also provides a blasting vibration prediction system based on an atomic search algorithm to optimize a BP neural network, comprising a memory and a processor, the memory being used to store executable program code; the processor being connected to the memory, and running a computer program corresponding to the executable program code by reading the executable program code, so as to execute any of the aforementioned blasting vibration prediction methods based on an atomic search algorithm to optimize a BP neural network.
[0015] The present application provides a blasting vibration prediction method based on an atomic search algorithm to optimize a BP neural network. The atomic search algorithm (ASO) is used to optimize the weights and thresholds of the BP neural network, thereby solving the problem that the standard BP network is prone to falling into local optimality and difficulty in adjusting parameters. The algorithm combines the ASO algorithm with the BP neural network, optimizes the network training process, and improves the generalization ability and prediction accuracy of the model. The algorithm can automatically learn and identify the laws of blasting vibration velocity, and can respond quickly even to nonlinear and time-varying complex sample parameters. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.
[0017] Figure 1 It is a flow chart of a method for predicting blasting vibration based on an atomic search algorithm and BP neural network optimization provided by the present application; Figure 2 It is the BP neural network model in the blasting vibration prediction method based on the atomic search algorithm to optimize the BP neural network provided in this application; Figure 3 This is a specific step of step S500 in the blasting vibration prediction method based on the atomic search algorithm to optimize the BP neural network provided in this application. DETAILED DESCRIPTION
[0018] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application. In addition, it should be understood that the specific implementation methods described herein are only used to illustrate and explain the present application and are not used to limit the present application. In the present application, unless otherwise stated, the directional words used, such as "up", "down", "left", and "right", generally refer to the up, down, left, and right of the device in actual use or working state, specifically the drawing direction in the accompanying drawings.
[0019] The present application provides a method for predicting blasting vibration based on an atomic search algorithm to optimize a BP neural network, which is described in detail below. It should be noted that the order of description of the following embodiments is not intended to limit the preferred order of the embodiments of the present application. In the following embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a particular embodiment, please refer to the relevant description of other embodiments.
[0020] See also Figure 1 The present application provides a method for predicting blasting vibration based on an atom search algorithm to optimize a BP neural network. The atom search optimization algorithm (ASO) is a new meta-heuristic global optimization method proposed by Zhao et al. based on atomic dynamics. The algorithm is a population-based iteration that imitates the atomic motion model in nature. It interacts through the interaction force generated by the Lennard-Jones (LJ) potential and the constraint force generated by the bond length potential, thereby continuously updating the position of the atom until the atom is in the optimal position. The algorithm iteration is completed. Due to the advantages of simplicity, fast convergence speed and strong optimization ability, the algorithm has been applied to many fields such as diffusion coefficient estimation in groundwater, hydrogeological parameter estimation, automatic clustering and fuel cell model parameter estimation.
[0021] See also Figure 1-Figure 3 The blasting vibration prediction method based on the atomic search algorithm to optimize the BP neural network specifically includes the following steps: S100, establishing a blasting vibration sample database, wherein the input sample data of the blasting vibration sample database includes blasting center distance, elevation difference, hole spacing, row spacing, hole depth, single hole charge, maximum section charge, delay time, minimum resistance line, and rock density; Please refer to Table 1. This application provides the deep hole step blasting parameters of the open pit of Zijinshan Gold and Copper Mine as the basis, and collects blasting vibration data on site for prediction. The mine step height is 12m, and the on-site mixed ammonium oil explosive is used. The continuous coupling charging structure, the blast hole diameter is 140mm, and the digital electronic detonator is used for hole-by-hole detonation network. Sichuan Zhongke Surveying and Mapping Company's TC-4850 blasting vibration meter is used for on-site monitoring to obtain 80 sets of sample data, among which the particle peak velocity (PPV), the maximum charge of a single section, the distance from the explosion center, the elevation difference and the unit consumption of explosives are used as the input parameters of the ASO-BP model.
[0022] Table 1 Part of the training sample data Serial number Maximum dosage per stage / kg Explosion center distance / m Elevation difference / m <![CDATA[Unit consumption of explosive / (kg / m 3 )]]> Particle peak velocity (PPV) / (cm / s) 1 115.4 56.5 0 0.49 4.21 2 180 122.2 12 0.56 2.89 3 180 102.8 12 0.54 4.09 4 162 170.86 12 0.54 0.38 5 175 162 12 0.52 0.67 6 175 166.9 12 0.54 0.45 7 149 137.7 0 0.55 1.99 8 137.5 68.1 0 0.52 4.26 9 137.5 84.6 0 0.52 3.34 10 137.5 92.9 12 0.52 2.36 11 137.5 107.7 12 0.52 0.90 12 137 144.8 12 0.52 1.05 13 137.5 76.7 0 0.52 2.19 … … … … … … S200, obtaining output result sample data, wherein the output result sample data includes the particle peak velocity (PPV) of the measured point in three directions of XYZ; S300, preprocessing the input sample data and the output result sample data; The preprocessing includes the following steps: The input sample data and the output result sample data are normalized: Among them, in the formula: is the normalized value, and the normalized data are all in the range of [-1,1]. is the value of the factor to be normalized, is the minimum value of this factor in the sample database, is the maximum value of this factor in the sample database.
[0023] The input and output sample data are normalized, the actual blasting vibration sample data are selected, 80% of the samples are randomly selected for training, and 20% of the samples are used for prediction.
[0024] Table 2 Some sample data after normalization Serial number Maximum dosage per stage / kg Explosion center distance / m Elevation difference / m <![CDATA[Specific charge of explosive / (kg / m 3 )]]> Particle peak velocity (PPV) / (cm / s) 1 -1 -1.00 -1 -1.00 0.97 2 1 0.15 1 1.00 0.29 3 1 -0.19 1 0.43 0.91 4 0.44 1.00 1 0.43 -1.00 5 0.85 0.85 1 -0.14 -0.85 6 0.85 0.93 1 0.43 -0.96 7 0.04 0.42 -1 0.71 -0.17 8 -0.32 -0.80 -1 -0.14 1.00 9 -0.32 -0.51 -1 -0.14 0.53 10 -0.32 -0.36 1 -0.14 0.02 11 -0.32 -0.10 1 -0.14 -0.73 12 -0.33 0.55 1 -0.14 -0.65 13 -0.32 -0.65 -1 -0.14 -0.07 … … … … … … S400, constructing a BP neural network, and obtaining the number of hidden layers and the number of neurons in the hidden layer; The BP neural network model is a multi-layer feedforward neural network trained based on the error back propagation algorithm. It has good nonlinear prediction capabilities. Its learning process consists of two processes: forward propagation and back propagation of signals. During forward propagation, the signal is calculated from the input layer, and the weighted sum of each layer is finally transmitted to the output layer through each hidden layer to obtain the output result. The output result is compared with the expected result to obtain the output error. Error back propagation is to propagate the error layer by layer from the hidden layer to the input layer according to the gradient descent algorithm, and distribute the error to all units in each layer, thereby obtaining the error signal of each unit, and modifying the weight of each unit accordingly. When the minimum error or maximum number of training times is met, the network training stops.
[0025] In this embodiment, the parameters are converted into mathematical methods according to the nonlinear mapping relationship of neurons for calculation without knowing the functional relationship between each group of blasting parameters and blasting vibration, and finally the predicted value is obtained. The BP network prediction needs to determine the number of hidden layers and the number of neurons in the hidden layer. The step of using a single hidden layer to construct a 3-layer neural network model according to the Kolmogorov theorem includes: According to Kolmogorov's theorem, given any continuous function f:{U}^{n}\to {R}^{m},f\left ( {x} \right )=y,U\in \left [ {0,1} \right ] , Indicates from A subset of 2-dimensional Euclidean space arrive dimensional Euclidean space subset The mapping of is the unit closed interval \left [ {0,1} \right ] , continuous function It is implemented using a 3-layer feedforward neural network. The input layer of this network has neurons, and the hidden layer has neurons, layer 3 (output layer), A neuron.
[0026] Therefore, according to Kolmogorov's theorem, a three-layer neural network model is constructed using a single hidden layer. The network structure of the three-layer neural network model includes an input layer, an output layer and a hidden layer. Figure 2 As shown, the first layer (input layer) of this network has processing units, and the middle layer (hidden layer) has processing units, and the third layer (output layer) has processing unit.
[0027] Determine the number of hidden layer nodes: in, is the number of nodes in the input layer; is the number of hidden layer nodes; is the number of nodes in the output layer; It is the adjustment constant between hidden layers and takes an integer between 1 and 10.
[0028] S500, optimize BP neural network based on ASO algorithm, build ASO-BP blasting vibration velocity prediction model, and determine network topology structure; S600 , inputting acquisition parameters into the ASO-BP blasting vibration velocity prediction model, and obtaining blasting vibration velocity prediction data.
[0029] BP neural network has strong nonlinear mapping ability, but uses gradient descent method to obtain error to reversely adjust the weights and thresholds of neurons. When facing relatively complex problems, the algorithm is prone to fall into the problem of local optimal solution and algorithm divergence. In order to optimize the problem of BP neural network falling into local minimum, the strong global optimization ability of ASO algorithm is introduced to optimize the weights and thresholds of the network model.
[0030] The ASO-BP blasting velocity prediction model was constructed based on MATLAB language. The main purpose was to determine the number of hidden layer nodes. The number of input layer nodes was 4 (i.e., the maximum charge per section, blast center distance, elevation difference, and unit explosive consumption), and the number of output layer nodes was one (i.e., PPV). The number of hidden layer nodes was determined by Kolmogorov theorem, and 10 different node numbers between [3, 12] were obtained. The number of nodes was selected by comparing the training error obtained by different node numbers each time. The BP neural network was trained 1 000 times, with a learning rate of 0.01 and a minimum error of 1×10 -6 , the maximum number of failures is 6 times; the initial atomic group size of the ASO algorithm is 30, the maximum number of iterations is 50, and the first 60 groups of the collected 80 data are selected as training sets for training, and the last 20 groups are used as test sets. Table 3 can be obtained through iterative calculation.
[0031] Table 3 Number of hidden layer nodes Average relative error / % 3 47.1 4 45.8 5 60.2 6 47.0 7 72.9 8 28.7 9 66.7 10 103.1 11 43.8 12 134.5 It can be seen from Table 3 that the error is the smallest when the number of hidden layer nodes is 8. Therefore, a network topology of (4, 8, 1) is built based on the collected sample data.
[0032] The steps of optimizing the BP neural network based on the ASO algorithm specifically include the following steps: S510, using the pre-processed input sample data and the output result sample data to perform initial network training on the BP neural network to obtain network weights and thresholds; The initial training steps are as follows: S511, in the interval \left [ {-1,1} \right ] Initialize the network weights randomly , , threshold 、 , Represents the input layer neurons and hidden layer The connection weights between neurons; represents the hidden layer The neurons in the output layer The connection weights between neurons; Represents the threshold of the hidden layer neurons ， represents the threshold of the output layer neuron; Indicates the number of input layer nodes; Represents the number of hidden layer nodes; , is the number of nodes in the output layer; S512, Assume that all hidden layer neurons use activation function function: , in, Represents the input signal of the hidden layer; The function is a hyperbolic tangent curve about x. When x approaches positive infinity or negative infinity, the function approaches 1 or -1, and when x is equal to 0, the value of the function is 0.
[0033] S513, calculate the output value of the hidden layer neuron ; , in, represents the hidden layer activation function; Indicates that the neuron receives The input signal passed from the input layer; Represents the input layer neurons and hidden layer The connection weights between neurons; Indicates Input signal, represents the hidden layer The threshold of each neuron; ， is the number of hidden layer nodes; S514, calculate the output value of the output layer neuron ; , in, represents the output value of the output layer neuron, represents the output layer activation function, ， is the number of hidden layer nodes, ， is the number of nodes in the output layer; represents the hidden layer neurons and the output layer The connection weights between neurons; represents the output value of the hidden layer neuron, Represents the output layer The threshold of each neuron; S515, define the error function as: , in, Indicates the number of training samples; is the number of training sample groups; ， is the number of nodes in the output layer, represents the predicted value of the output layer, Indicates actual value; S516, update the connection weight to: , in, is the learning rate parameter, in the interval \left [ {0,1} \right ] Take values between Indicates At the iteration, the hidden layer The neurons in the output layer The connection weights between neurons; represents the output value of the hidden layer neuron, , is the number of nodes in the output layer, Indicates Input signal, represents the output value of the output layer neuron, represents the predicted value of the output layer, Indicates the number of iterations; S517, update threshold value: , in, Indicates At the iteration, the hidden layer The threshold of each neuron; The output layer The threshold of the neuron; is the learning rate parameter, in the interval \left [ {0,1} \right ] Take values between represents the output value of the hidden layer neuron, Indicates Input signal, represents the hidden layer The neurons in the output layer The connection weights between neurons; represents the output value of the output layer neuron, represents the predicted value of the output layer neuron, Indicates the number of iterations; S518, determining whether the iteration is finished, if not, returning to the step of calculating the output value of the hidden layer neuron; The condition for the iteration to end is that the preset maximum number of iterations is reached or the fitness value reaches a preset threshold.
[0034] S520, performing real-value encoding on network weights and thresholds to form initial individuals, and obtaining an initial atomic population; S530, taking the training error of the BP neural network as the individual fitness value, that is, selecting the mean absolute error of the training set and the test set as the fitness of ASO, and obtaining the fitness function: , \left [ {0,1} \right ] in, is the expected output value of the BP neural network, is the actual output value of the BP neural network, - represents the training error of BP neural network, represents the sample size of the training set, express Function to get the dimension of an array. Represents the fitness value, which is the standard for evaluating the quality of individuals; S540, using the ASO algorithm to update the atomic population and using the fitness function to calculate the fitness values of individuals in the updated atomic population; S550, by iteratively calculating the best fitness of each atom, selecting the best individual, and optimizing the BP neural network, wherein the first Dai Zaiwei Previous The acceleration of an atom is: in, Indicates The acceleration of atoms, Indicates The interaction force on each atom is Indicates The covalent bond forces on atoms Indicates The mass of an atom; No. In the iteration The mass of an atom It is determined by the fitness value of the individual atomic group, and its definition is as follows: , Where: Indicates The estimated mass of atoms, and Respectively indicate the The objective function fitness values of the best and worst atoms during iteration; Indicates Atom in The function fitness value at the iteration; represents the total number of atoms, Indicates atoms.
[0035] To simplify the ASO algorithm, in the global optimization process of the algorithm, at the t+1th iteration i The position and velocity of an atom in the dth dimension are expressed as: , in, Indicates Generation The first The speed of atoms, Represents the interval \left [ {0,1} \right ] A random number between Indicates Generation Wei Shangdi The speed of atoms, Indicates Generation Wei Shangdi The acceleration of atoms, Indicates Generation Wei Shangdi The position of atoms, Indicates Generation Wei Shangdi The position of atoms; S560 and BP neural networks are trained and tested using weights and thresholds optimized by the APO algorithm.
[0036] When training the network model, the weights and thresholds of the BP neural network are set as the atomic population parameters of the ASO algorithm, and the prediction error of the BP neural network is used as the fitness value of the ASO algorithm. The global optimal solution is found according to the steps of the algorithm, and it is used as the weights and thresholds of the BP neural network. The optimized weights and thresholds are used as the initial values of the BP neural network, and data training is performed on it to obtain the corresponding neural network model.
[0037] In step S550, obtain the The interaction forces on atoms The solution formula is: , Where: interaction force Indicates the surrounding atoms The sum of the forces acting on the atoms is represents the current search space dimension, ; Indicates the current iteration number; is a subset of the total number of atoms, indicating The set of atoms with the best fitness function value; is the interval \left [ {0,1} \right ] A random number between Indicates In the iteration, Atomic pair Lennard-Jones potential force of atoms; The Lennard-Jones potential force is expressed as: ,, Where: interaction force Indicates the surrounding atoms The sum of the forces acting on the atoms is is a depth function used to adjust the repulsive and attractive areas. For the Atoms and The scaled distance between atoms, Indicates the current iteration number; is the depth function, as shown below: Where: is the depth weight, is the maximum number of iterations, Indicates the current iteration number; For the Atoms and The scale distance between atoms is different The value can be understood as the different force properties of atoms. When the value is between 0.9 and 1.12, rejection occurs. Attraction occurs at 1.12~2, and when Equilibrium occurs when ) begins, and the attraction increases with The increase first gradually increases and reaches a maximum value ( ), and then gradually decrease until When , the attraction is approximately equal to zero. Therefore, in order to ensure the global effectiveness of the ASO algorithm, the lower limit of the repulsive force with a smaller function value is set to , the upper limit of attraction with a larger function value is set to , which can be defined as: , Where: and They are The lower and upper limits of and The definition of is as follows: , Where: , Respectively represent the Atoms and The position of atoms; is the Euclidean distance between atoms; Indicates the current iteration number; , Where: , Respectively represent the Atoms and The position of atoms; represents the collision scale, is a subset of the total number of atoms, indicating The set of atoms with the best fitness function value, Indicates the current iteration number; is a subset of the total number of atoms, indicating The set of atoms with the best fitness function value. During the iteration process, atoms interact with fewer and fewer neighboring atoms to enhance the local development ability of the algorithm and ensure the convergence of the algorithm.
[0038] is a time function, which decreases as the number of iterations increases. The definition of is as follows: Where: is the total number of atoms; is the total number of iterations, Indicates the current iteration number.
[0039] Therefore, the interaction force It can describe the motion of simple molecules. For more complex molecules or atomic systems, it is necessary to introduce a geometrically constrained molecular dynamics method and combine it with the internal motion of atoms to simulate the atomic optimization process.
[0040] In the ASO algorithm, it is assumed that each atom has a covalent bond with the best atom and each atom is constrained by the best atom. The atoms are constrained to , Where: It is The position of the best atom at iteration ; It is The square of the fixed bond length between the atom and the optimal atom, then The binding force of atoms is , Where: Represents the current search space dimension, Indicates the number of times in the current search space dimension Iteration Atomic constraints, is the dimension of the current search space. The best atomic position of the iteration, is the dimension of the current search space. Atom The position of the iteration, is the Lagrange multiplier, , is the multiplier weight, is the total number of iterations; let , then the constraint force caused by the covalent bond is solved as follows: , Where: is the multiplier weight, is the best atomic position for the tth iteration in the current search space dimension, is the dimension of the current search space. Atom The position of the iteration, is the total number of iterations, so under the interaction force and geometric constraints, Atoms in The acceleration at this moment is: , in, Indicates The acceleration of atoms, Indicates The interaction force on each atom is Indicates The covalent bond forces on atoms Indicates The mass of an atom, is the depth weight, Indicates the current iteration number, is the total number of iterations, represents the current search space dimension, , is a subset of the total number of atoms, indicating The set of atoms with the best fitness function value; is the interval \left [ {0,1} \right ] A random number between Indicates In the iteration, Atomic pair The Lennard-Jones potential force of atoms, is the dimension of the current search space. Atom The position of the iteration, is the dimension of the current search space. Atom The position of the iteration, For the The best atomic position of the iteration, is the multiplier weight.
[0041] The atomic search algorithm (ASO) is used to optimize the weights and thresholds of the BP neural network, solving the problem that the standard BP network is prone to falling into local optimality and difficulty in adjusting parameters. This is one of the key technologies of this patent.
[0042] The ASO-BP blasting vibration velocity prediction model combines the ASO algorithm with the BP neural network, optimizes the network training process, and improves the generalization ability and prediction accuracy of the model. The algorithm can automatically learn and identify the laws of blasting vibration velocity, and can respond quickly even to nonlinear and time-varying complex sample parameters.
[0043] The present application also provides a blasting vibration prediction system based on an atomic search algorithm to optimize a BP neural network, comprising a memory and a processor, the memory being used to store executable program code; the processor being connected to the memory, and running a computer program corresponding to the executable program code by reading the executable program code, so as to execute any of the aforementioned blasting vibration prediction methods based on an atomic search algorithm to optimize a BP neural network.
[0044] The above is a detailed introduction to the blasting vibration prediction method based on the atomic search algorithm to optimize the BP neural network provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for general technical personnel in this field, according to the idea of the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.
Claims
1. A blasting vibration prediction method based on BP neural network optimized by atomic search algorithm is characterized by: include: Establishing a blasting vibration sample database, wherein the input sample data of the blasting vibration sample database includes blasting center distance, elevation difference, hole spacing, row spacing, hole depth, single hole charge, maximum section charge, delay time, minimum resistance line, and rock density; Obtaining output result sample data, wherein the output result sample data includes the peak vibration velocity of the particle at the measured point in three directions of XYZ; Preprocessing the input sample data and the output result sample data; Construct a BP neural network and obtain the number of hidden layers and neurons in the hidden layer; Based on the ASO algorithm, the BP neural network is optimized, the ASO-BP blasting vibration velocity prediction model is constructed, and the network topology is determined; Input the acquisition parameters into the ASO-BP blasting vibration velocity prediction model and obtain the blasting vibration velocity prediction data.
2. The blasting vibration prediction method based on BP neural network optimized by atomic search algorithm according to claim 1 is characterized in that: Preprocessing the input sample data and the output result sample data includes: The input sample data and the output result sample data are normalized: Among them, in the formula: is the normalized value, and the normalized data are all in the range of [-1,1]. is the value of the factor to be normalized, is the minimum value of this factor in the sample database, is the maximum value of this factor in the sample database.
3. The blasting vibration prediction method based on BP neural network optimized by atomic search algorithm according to claim 1 is characterized in that: Construct a BP neural network and obtain the number of hidden layers and neurons in the hidden layer, including: According to Kolmogorov theorem, a three-layer neural network model is constructed using a single hidden layer, and the network structure of the three-layer neural network model includes an input layer, an output layer and a hidden layer; Determine the number of hidden layer nodes: in, is the number of nodes in the input layer; is the number of hidden layer nodes; is the number of nodes in the output layer; It is the adjustment constant between hidden layers and takes an integer between 1 and 10.
4. The blasting vibration prediction method based on BP neural network optimized by atomic search algorithm according to claim 3 is characterized in that: According to Kolmogorov's theorem, a three-layer neural network model is constructed using a single hidden layer, including: According to Kolmogorov's theorem, given any continuous function , Indicates from A subset of the 2-dimensional Euclidean space arrive dimensional Euclidean space subset The mapping of is a unit closed interval , continuous function It is implemented using a 3-layer feedforward neural network. The input layer of this network has neurons, and the hidden layer has neurons, and the third layer, the output layer, has A neuron.
5. The blasting vibration prediction method based on BP neural network optimized by atomic search algorithm according to claim 1 is characterized in that: Optimize BP neural network based on ASO algorithm, including: Using the pre-processed input sample data and the output result sample data to perform initial network training on the BP neural network to obtain network weights and thresholds; The network weights and thresholds are encoded with real values to form initial individuals and obtain the initial atomic population; The training error of the BP neural network is used as the individual fitness value, that is, the mean absolute error of the training set and the test set is selected as the fitness of ASO, and the fitness function is obtained: ， in, is the expected output value of the BP neural network, is the actual output value of the BP neural network, - represents the training error of BP neural network, represents the sample size of the training set, express Function to get the dimension of an array. Represents the fitness value, which is the standard for evaluating the quality of individuals; The ASO algorithm is used to update the atomic population and the fitness function is used to calculate the fitness value of individuals in the updated atomic population; By iteratively calculating the best fitness of each atom, selecting the best individual, and optimizing the BP neural network, Dai Zaiwei Previous The acceleration of an atom is: in, Indicates The acceleration of atoms, Indicates The interaction force on each atom is Indicates The covalent bond forces on atoms Indicates The mass of an atom; No. In the iteration The mass of an atom It is determined by the fitness value of the individual atomic group, and its definition is as follows: ， Where: Indicates The estimated mass of atoms, and Respectively indicate the The objective function fitness values of the best and worst atoms during iteration; Indicates Atom in The function fitness value at the iteration; represents the total number of atoms, Indicates atoms, To simplify the ASO algorithm, in the global optimization process of the algorithm, at the t+1th iteration i The position and velocity of an atom in the dth dimension are expressed as: ， in, Indicates Generation The first The speed of atoms, Representation interval A random number between Indicates Generation Wei Shangdi The speed of atoms, Indicates Generation Wei Shangdi The acceleration of atoms, Indicates Generation Wei Shangdi The position of atoms, Indicates Generation Wei Shangdi The position of atoms; The BP neural network is trained and tested using weights and thresholds optimized by the APO algorithm.
6. The blasting vibration prediction method based on BP neural network optimized by atomic search algorithm according to claim 5 is characterized in that: By iteratively calculating the best fitness of each atom, selecting the best individual, and optimizing the BP neural network, Get the The interaction forces on atoms The solution formula is: ， Where: interaction force Indicates the surrounding atoms The sum of the forces acting on the atoms is represents the current search space dimension, ; Indicates the current iteration number; is a subset of the total number of atoms, indicating The set of atoms with the best fitness function value; For interval A random number between Indicates In the iteration, Atomic pair Lennard-Jones potential force of atoms; The Lennard-Jones potential force is expressed as: ， Where: interaction force Indicates the surrounding atoms The sum of the forces acting on the atoms is is a depth function used to adjust the repulsive and attractive areas. For the Atoms and The scaled distance between atoms, Indicates the current iteration number; is the depth function, as shown below: Where: is the depth weight, is the maximum number of iterations, Indicates the current iteration number; For the Atoms and The scaled distance between atoms is set to , the upper limit of attraction with a larger function value is set to , which is defined as: , Where: and They are The lower and upper limits of and The definition of is as follows: ， Where: , Respectively represent the Atoms and The position of atoms; is the Euclidean distance between atoms; Indicates the current iteration number; ， Where: , Respectively represent the Atoms and The position of atoms; represents the collision scale, is a subset of the total number of atoms, indicating The set of atoms with the best fitness function value, Indicates the current iteration number; is a time function, which decreases as the number of iterations increases. The definition of is as follows: Where: is the total number of atoms; is the total number of iterations, Indicates the current iteration number.
7. The method for blasting vibration prediction based on BP neural network optimized by atomic search algorithm according to claim 6 is characterized in that: By iteratively calculating the best fitness of each atom, selecting the best individual, and optimizing the BP neural network, Assuming that each atom has a covalent bond with the best atom and each atom is bound by the best atom, The atoms are constrained to ， Where: It is The position of the best atom at iteration ; It is The square of the fixed bond length between the atom and the optimal atom, then The binding force of atoms is ， Where: Represents the current search space dimension, Indicates the number of times in the current search space dimension Iteration Atomic constraints, is the dimension of the current search space. The best atomic position of the iteration, is the dimension of the current search space. Atom The position of the iteration, is the Lagrange multiplier, , is the multiplier weight, is the total number of iterations; let , then the constraint force caused by the covalent bond is solved as follows: ， Where: is the multiplier weight, is the best atomic position for the tth iteration in the current search space dimension, is the dimension of the current search space. Atom The position of the iteration, is the total number of iterations, so under the interaction force and geometric constraints, Atoms in The acceleration at this moment is: ， in, Indicates The acceleration of atoms, Indicates The interaction force on each atom is Indicates The covalent bond forces on atoms Indicates The mass of an atom, is the depth weight, Indicates the current iteration number, is the total number of iterations, represents the current search space dimension, , is a subset of the total number of atoms, indicating The set of atoms with the best fitness function value; For interval A random number between Indicates In the iteration, Atomic pair The Lennard-Jones potential force of atoms, is the dimension of the current search space. Atom The position of the iteration, is the dimension of the current search space. Atom The position of the iteration, For the The best atomic position of the iteration, is the multiplier weight.
8. The method for blasting vibration prediction based on BP neural network optimized by atomic search algorithm according to claim 5 is characterized in that: Using the pre-processed input sample data and the output result sample data to perform initial network training on the BP neural network to obtain network weights and thresholds, including: In the interval Initialize the network weights randomly , , threshold 、 , Represents the input layer neurons and hidden layer The connection weights between neurons; represents the hidden layer The neurons in the output layer The connection weights between neurons; Represents the threshold of the hidden layer neurons ， represents the threshold of the output layer neuron; Indicates the number of input layer nodes; Represents the number of hidden layer nodes; , is the number of nodes in the output layer; Assume that all hidden layer neurons use the activation function tan sig function: ， in, Represents the input signal of the hidden layer; Calculate the output value of the hidden layer neurons ; ， in, represents the hidden layer activation function; Indicates that the neuron receives The input signal passed from the input layer; Represents the input layer neurons and hidden layer The connection weights between neurons; Indicates Input signal, represents the hidden layer The threshold of each neuron; ， is the number of hidden layer nodes; Calculate the output value of the output layer neuron ; ， in, represents the output value of the output layer neuron, represents the output layer activation function, ， is the number of hidden layer nodes, ， is the number of nodes in the output layer; represents the hidden layer neurons and the output layer The connection weights between neurons; represents the output value of the hidden layer neuron, Represents the output layer The threshold of each neuron; The error function is defined as: ， in, Indicates the number of training samples; is the number of training sample groups; ， is the number of nodes in the output layer, represents the predicted value of the output layer, Indicates actual value; Update the connection weight to: ， in, is the learning rate parameter, in the interval Take values between Indicates At the iteration, the hidden layer The neurons in the output layer The connection weights between neurons; represents the output value of the hidden layer neuron, , is the number of nodes in the output layer, Indicates Input signal, represents the output value of the output layer neuron, represents the predicted value of the output layer, Indicates the number of iterations; Update thresholds: ， in, Indicates At the iteration, the hidden layer The threshold of each neuron; The output layer The threshold of the neuron; is the learning rate parameter, in the interval Take values between represents the output value of the hidden layer neuron, Indicates Input signal, represents the hidden layer The neurons in the output layer The connection weights between neurons; represents the output value of the output layer neuron, represents the predicted value of the output layer neuron, Indicates the number of iterations; Determine whether the iteration is finished. If not, return to the step of calculating the output value of the hidden layer neuron.
9. The method for blasting vibration prediction based on BP neural network optimized by atomic search algorithm according to claim 8 is characterized in that: In determining whether the iteration is finished, the condition for the iteration to end is that a preset maximum number of iterations is reached or the fitness value reaches a preset threshold.
10. The blasting vibration prediction system based on the BP neural network optimized by the atomic search algorithm is characterized by: include: A memory for storing executable program codes; as well as A processor is connected to the memory, and runs a computer program corresponding to the executable program code by reading the executable program code, so as to execute the blasting vibration prediction method based on the atomic search algorithm to optimize the BP neural network as described in any one of claims 1 to 9.
Citation Information
Cited By
Settlement-control-oriented dynamic control method and system for vibration of station change project
CN121785139A