A neural network-based intelligent evaluation method for mine rock mass explosibility zoning

Abstract

The application discloses a kind of based on neural network's mine rock mass explodability partition intelligent evaluation method, the method includes the following steps: step S1, the generation of data set;Step S2: the establishment of BP neural network;Step S3, the setting of genetic algorithm;Step S4, the determination of genetic operation;Step S5, initial weight and threshold optimization;Step S6, the test and adjustment of BP neural network;Step S7, the collection of mine blasting index data;Step S8, mine rock mass explodability partition intelligent evaluation.Compared with the traditional mine rock mass explodability partition method, the neural network can be used to more accurately evaluate the mine rock mass explodability partition, effectively improve the evaluation efficiency.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of blasting engineering, and relates to a mine rock mass explosibility zoning evaluation method, in particular to a mine rock mass explosibility zoning intelligent evaluation method based on a neural network. BACKGROUND

[0002] Rock mass explosibility classification reflects the difficulty of rock mass destruction under the action of explosive explosion, and can also be understood as the ability of rock to resist destruction under the action of explosive explosion. The traditional research object of rock mass explosibility classification is rock. Mine blasting often faces problems such as complex mine geological structure, multiple types of ore and rock, and interbedded symbiosis of ore and rock. Especially for large-diameter bench blasting production, large hole pattern parameters, and large differences in lithology between two holes, therefore, the simple explosibility classification of rock as the research object is not suitable for complex ore body mining area with multiple ore types, and cannot comprehensively reflect the damage characteristics of blasting on ore and rock and the advantages and disadvantages of blasting quality.

[0003] Rock mass blasting zoning is to divide the area with equivalent or consistent explosibility into a blasting zone, determine reasonable hole pattern parameters for different blasting zones, adjust the charge quantity, improve the charge structure, optimize the blasting network, and improve the blasting effect. Blasting zoning is the development and improvement of traditional rock mass explosibility classification, so as to better adapt to the needs of efficient mining of complex ore body with multiple ore types in non-coal mines. Mines organize perforation blasting by taking the blasting zone as the unit, therefore, the explosibility zoning of rock mass in the mine stope is an inevitable choice for blasting technology optimization, improvement of blasting effect, improvement of blasting mining efficiency and safety production.

[0004] At present, the explosibility zoning evaluation of complex rock mass in mine area mostly adopts multi-index comprehensive evaluation method, such as grey clustering analysis, matter element analysis model and attribute recognition model, which has problems of lack of pertinence, certain subjectivity and low evaluation efficiency. SUMMARY

[0005] In order to solve the above problems in the background art, the application provides a mine rock mass explosibility zoning intelligent evaluation method based on a neural network.

[0006] The purpose of the application is achieved by the following technical scheme:

[0007] A mine rock mass explosibility zoning intelligent evaluation method based on a neural network comprises the following steps:

[0008] Step S1, generation of data set:

[0009] Step S101, data set generation: according to the summary of existing literature, the standard of rock mass blastability zoning is summarized, the partition interval value of each single evaluation factor corresponding to each rock mass blastability grade, the input layer corresponding to each rock mass blastability grade is randomly valued in the index limit range by using the round and rand functions in excel, and sample data is generated;

[0010] Step S102, data set standardization: the data set is normalized to obtain normalized data;

[0011] Step S103, data set division: the data set is divided into training set and test set;

[0012] Step S2: establishment of BP neural network

[0013] Select rock density, uniaxial compressive strength of rock, uniaxial tensile strength of rock, rock integrity coefficient, average block size, and large block rate as the input layer of BP neural network, and divide the rock mass blastability zoning into five grades of extremely easy to blast, easy to blast, medium, difficult to blast, and extremely difficult to blast as the output layer of BP neural network. The specific establishment steps are as follows:

[0014] Step S201, determination of the number of hidden layer neurons:

[0015] The number of hidden layer neurons is determined by the following formula:

[0016]

[0017] Wherein, k is the number of hidden layer neurons; e is the number of input layer; u is the number of output layer; a is a constant between 1 and 10;

[0018] Step S202, determination of BP neural network activation function and training function:

[0019] According to the needs, the hidden layer activation function, the output layer activation function and the training function are selected;

[0020] Step S203, determination of other parameters of BP neural network:

[0021] According to the needs, the maximum learning times, the initial learning rate, the initial momentum factor and the training accuracy of BP neural network are selected;

[0022] Step S3, setting of genetic algorithm:

[0023] Step S301, determination of coding range and length: the weight and threshold value range of BP neural network is between [-1, 1], the coding range of chromosome in genetic algorithm corresponds to the weight and threshold value of BP neural network, and the coding length calculation formula is as follows:

[0024] l = e*k + k*u + (e + u)

[0025] Wherein, l is the encoding length, k is the number of hidden layer neurons; e is the number of input layer; u is the number of output layer;

[0026] Step S302, the determination of the encoding mode: using real number encoding method, the encoding value is a random number in the encoding range;

[0027] Step S303, the determination of the adaptive function: the weight and threshold information in the individual is assigned to the BP neural network, the BP neural network is trained using the training data set, and the fitness calculation formula is as follows:

[0028]

[0029] Wherein, F is the fitness, t i is the true value of the ith sample, y i is the predicted value of the ith sample; n is the number of training set samples;

[0030] Step S4, the determination of the genetic operation:

[0031] Step S401, the determination of the selection operation: the roulette algorithm is used for the selection operation of the genetic operation, and the formula is:

[0032]

[0033] Wherein, F i is the fitness of each individual, N is the population size, k is the number of hidden layer neurons, f i is the product of k and the reciprocal value of the fitness, P i is the selection probability of each individual;

[0034] Step S402, the determination of the crossover operation: the real number fusion crossover method is selected, the encoding of two individuals in a random position meeting the crossover probability is randomly fused to generate two new individuals, and the crossover operation formula is as follows:

[0035] a mk = a m (1-b) + a n b

[0036] a nk = a n (1-b) + a m b

[0037] Wherein, a mk and a nk are the mth individual a m and the nth individual a nThe generated individual is crossed at a random position k; b is a random number between 0 and 1;

[0038] Step S403, determination of the mutation operation: a real number mutation is adopted, and the formula is as follows:

[0039]

[0040] In the formula, a wk is the wth individual a w The generated individual is mutated at a random position k; a max is the upper bound of coding; a min is the lower bound of coding; f(g) = b(1-num / maxGen) 2 ; b and r are random numbers between 0 and 1; num is the current iteration number; maxGen is the preset maximum iteration number;

[0041] Step S404, determination of other parameters: the iteration number, population size, crossover probability and mutation probability need to be set before the genetic algorithm starts;

[0042] Step S5, initial weight and threshold optimization:

[0043] The initial weights and thresholds of the network are optimized by the genetic algorithm. After the global optimal solution is calculated by the genetic algorithm, the weights and thresholds are assigned to the BP neural network. The specific steps are as follows:

[0044] Step S501, population initialization: according to the structure of the BP neural network, the original population is randomly generated;

[0045] Step S502, chromosome coding: the chromosome is the weight and threshold of the BP neural network, and the chromosome is generated in the order of weight first and threshold last;

[0046] Step S503, the training set A is substituted into the BP neural network with the initialized weights and thresholds as model training data, and the error value of the BP neural network is obtained;

[0047] Step S504, individual fitness calculation: the error value obtained in step S503 is substituted into the fitness function to calculate the individual fitness and population average fitness of each individual;

[0048] Step S505, selection and replication: the individuals are sequentially arranged according to the fitness, and in this evolution process, the individuals located at the back are removed according to the roulette algorithm, and the individuals located at the front are copied for evolution;

[0049] Step S506, crossover and mutation: the population individuals are crossed and mutated according to the crossover and mutation formula;

[0050] Step S507, calculate the fitness value, determine whether the individual completing inheritance evolution is superior to the average fitness of the last generation, if the determination meets the set requirement, continue to step S508 to complete decoding, if not, go to step S505 for next cycle;

[0051] Step S508, decoding: decode the final winning chromosome according to the sorting and coding mode corresponding to step S502;

[0052] Step S509: assignment: according to step S508, the conversion of chromosome string to specific numerical value has been completed, and the set of weights and threshold values selected by genetic algorithm are delivered to BP neural network;

[0053] Step S6, test and optimization of BP neural network:

[0054] Step S601: input the test set B as the input parameter into the BP neural network with optimized weight and threshold value;

[0055] Step S602, update the weight and threshold value based on the error calculated by BP neural network;

[0056] Step S603, calculate the error, if the requirement is met, get the optimal weight and threshold value of BP neural network;

[0057] Step S7, collection of mine blasting index data:

[0058] Step S701, measure the rock mass density;

[0059] Step S702, measure the uniaxial compressive strength and tensile strength of rock;

[0060] Step S703, calculate the rock mass integrity coefficient K v :

[0061]

[0062] In the formula, V pm is the longitudinal wave velocity of rock mass, V pr is the longitudinal wave velocity of rock;

[0063] Step S704, calculate the average blockiness and large block rate;

[0064] Step S8, intelligent evaluation of mine rock mass explosibility zoning:

[0065] Input the six rock indexes obtained by the test into the BP neural network, output five values corresponding to the five explosibility zoning of mine rock mass, the value at the corresponding explosibility zoning level is 1, and the value at other explosibility zoning levels is 0.

[0066] Compared with the prior art, the present application has the following advantages:

[0067] 1. The BP neural network is prone to fall into a local solution due to the difference of data samples, and the network shows different performances in processing the same problem due to the randomness of the initial weight value threshold, the present application adopts the genetic algorithm to obtain the optimal BP neural network weight value and threshold, avoiding the above problems.

[0068] 2. Compared with the traditional mine rock mass blastability zoning method, the neural network can more accurately evaluate the mine rock mass blastability zoning, and effectively improve the evaluation efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0069] Figure 1 It is a flow chart of the intelligent evaluation method of the mine rock mass blastability zoning based on the neural network;

[0070] Figure 2 It is a genetic algorithm flow chart. DETAILED DESCRIPTION

[0071] The technical scheme of the present application will be further described below in conjunction with the drawings, but is not limited thereto, any modification or equivalent replacement to the technical scheme of the present application without departing from the spirit and scope of the present application shall be covered in the protection scope of the present application.

[0072] Selection of the mine rock mass blastability zoning index:

[0073] Physical property index: the physical properties of the rock related to blasting mainly include the mineral composition and composition characteristics of the rock, the density, porosity, wave impedance, weathering degree of the rock, etc. If the stress generated by the explosion of the explosive can overcome the internal resistance and inertia force of the rock mass, part of the energy of the explosion will be converted into the kinetic energy of the rock, causing the rock mass to displace or even throw. The density of the rock is generally related to the mineral composition and structure of the rock, the greater the density of the rock, the greater the energy consumed for breaking, moving and throwing the rock, and the worse the blastability of the rock mass. The physical index of rock density can be used to measure the difficulty of rock mass blasting.

[0074] Mechanical property index of the mine rock: rock mass blasting is essentially a mechanical behavior process involving energy conversion. The rock around the explosive column is affected by the blasting impact stress field, and the amplitude has the characteristics of space-time distribution. When the stress value reaches the rock strength, part of the energy released by the explosion of the explosive will be consumed by the rock, which will cause the internal stress of the rock mass to decrease. The rock strength index can reflect the ability of the rock to resist damage under external force, and the greater the rock strength, the more difficult it is to be broken. Therefore, the uniaxial compressive strength and tensile strength of the rock are important indexes for characterizing the mechanical properties and deformation characteristics of the rock.

[0075] Fracture index of ore rock: Most blasting operations regard the rock mass as an aggregation of rock blocks, and there are certain structural planes, such as joints and fractures, between the blocks. The development of the structure of the natural rock mass is closely related to the final blasting effect. In this case, the main role of blasting is to cause the rock mass to break along the structural planes, resulting in a certain degree of fracture of the rock blocks, so that they become smaller rocks. Therefore, the more developed the geological structure of the rock mass, the higher the degree of fragmentation of the rock mass before blasting, and the less energy required to achieve the desired blasting fragmentation effect, that is, the better the blastability of the rock mass. Therefore, when evaluating the blastability of rock, the development of the structure and structure of the rock mass must be considered.

[0076] Blasting quality index of ore rock: Blasting size is a comprehensive quality index for evaluating blasting effect, which depends on the mechanical properties, fracture, explosive, blasting parameters and blasting process of ore rock. When the strength of the ore rock is large and the integrity is good, the energy consumption per unit of broken ore rock is large, and the ore rock is difficult to break, showing the characteristics of large blasting size and high large block rate. Therefore, the average size K 50 and the large block rate are selected as the blasting quality index for zoning. Generally speaking, the larger K 50 and the higher the large block rate, the more difficult the ore rock is to blast.

[0077] Backpropagation Neural Network (BP) is a commonly used artificial neural network model. The basic principle is to transmit and process information through the connection and weighting between multiple neurons, and continuously adjust the connection weights between neurons through the error backpropagation algorithm to achieve learning and prediction of input data. BP neural network is composed of input layer, hidden layer and output layer, and the hidden layer and output layer contain multiple neurons. The hidden layer performs weighted summation on the input data and nonlinear transformation through the activation function, and then transmits the results to the output layer.

[0078] Genetic Algorithm (GA) is a kind of simulation of Darwin's biological evolution theory, combined with natural selection and genetics theory, global search optimal solution. Core operations include: fitness calculation, selection, crossover and mutation. The specific process of genetic algorithm is: first, determine the appropriate coding scheme, initialize the coding to generate the original population P(t), evaluate the fitness of individuals in the population through the fitness function, and generate the fitness value Fitness(i). Fitness represents the advantages and disadvantages of each individual in its population, and the probability of individual inheritance to the next generation population is determined according to the size of fitness in the selection operation. The higher the fitness, the greater the possibility of individual inheritance to the next generation population. On the contrary, the smaller the probability of inheritance, the number of individuals inherited to the next generation population is determined by proportional selection, thereby forming a new population P(t+1). According to the biological evolution theory, when the individuals of the new population meet the crossover rate, the random crossover operation will occur. Through the crossover operation, two random individuals exchange the corresponding random gene fragments on the chromosome to produce new individuals. The individuals in the population have a certain mutation probability. When the individual reaches the mutation rate, the random gene on the chromosome will mutate, so that the individual produces a new genotype. After completing the above three operations, the fitness of the individuals in this round population is evaluated again, and the three operations are repeated to optimize the individuals of each generation, and stop after reaching the preset iteration number. Finally, select an individual with the highest fitness as the best solution to the problem.

[0079] According to the above principle, the application provides a neural network-based intelligent evaluation method for rock mass explosibility zoning in a mining area, as shown in Figure 1 The method comprises the following steps:

[0080] Step S1, generation of data set:

[0081] According to the summary of the existing literature, Table 1 shows the standard of rock mass explosibility classification:

[0082] Table 1 Standard of rock mass explosibility classification

[0083]

[0084] Step S101, data set generation: as shown in Table 1, the classification limit of each single evaluation factor corresponding to each rock mass explosibility grade is an interval value, and the input layer corresponding to each rock mass explosibility grade is randomly selected within the index limit range by using the round and rand functions in excel, and 500 sample data are generated.

[0085] Step S102, data set standardization: the data set is normalized to obtain normalized data.

[0086] In this step, the original data is processed by deviation standardization (min-max), and the standardization formula is as follows:

[0087]

[0088] In the formula, is the maximum value of the jth index; is the minimum value of the jth index;a ij is the ith element of the current index;b ij is the normalized value, ranging between [0, 1].

[0089] Step S103, data set division: divide the data set into training set A and test set B, and the sample proportions are 70% and 30%, respectively.

[0090] Step S2, establishment of BP neural network:

[0091] According to the above principle, six indexes of rock density, rock uniaxial compressive strength, tensile strength, rock mass integrity coefficient, average block size, and large block rate are selected, and the rock mass blastability is divided into five grades of extremely easy to blast, easy to blast, medium, difficult to blast, and extremely difficult to blast. Therefore, the number of input layer of the neural network model is 6, and the number of output layer is 5.

[0092] Step S201, determination of the number of hidden layer neurons: the number of neurons in the hidden layer is more complex to select, and its number is closely related to the required precision of error, the number of input and output units. Generally, the following formula is used to determine:

[0093]

[0094] Where, k is the number of hidden layer neurons; e is the number of input layer; u is the number of output layer; a is a constant between 1 and 10.

[0095] In this step, after repeated debugging, the number of hidden layer neurons is determined to be 12.

[0096] Step S202, determination of BP neural network activation function and training function: select the hidden layer activation function, output layer activation function and training function according to the needs.

[0097] In this step, the Logsig function is selected as the hidden layer activation function, the Purelin function is selected as the output layer activation function, and the gradient descent BP algorithm training function traingdx with momentum backpropagation and dynamic adaptive learning rate is selected as the training function.

[0098] Step S203, determination of other parameters of BP neural network: select the maximum learning times, initial learning rate, initial momentum factor and training accuracy of BP neural network according to the needs.

[0099] In this step, the maximum number of learning of the BP neural network is set to 500 times, the initial learning rate is set to 0.01, the initial momentum factor is set to 0.03, and the training requirement accuracy is set to 0.00001.

[0100] Step S3, setting of the genetic algorithm:

[0101] Step S301, determination of the coding range and length: the weight and threshold value range of the BP neural network is between [-1, 1], the coding range of the chromosome in the genetic algorithm corresponds to the weight and threshold value of the BP neural network, and the coding length calculation formula is as follows:

[0102] l = e·k + k·u + (e+u)

[0103] In the formula, l is the coding length, k is the number of hidden layer neurons; e is the number of input layer; and u is the number of output layer.

[0104] Step S302, determination of the coding method: the main coding methods include binary coding and real number coding. Since the binary coding cannot well reflect the structural characteristics and related information of the problem, and the randomness of the genetic algorithm makes the local search ability of the binary coding poor, and there is a distance problem between adjacent integers, the real number coding method is adopted here, and the coding value is a random number in the coding range.

[0105] Step S303, determination of the fitness function: the weight and threshold value information in the individual is assigned to the BP neural network, the BP neural network is trained using the training data set, and the fitness calculation formula is as shown in the following formula:

[0106]

[0107] In the formula, F is the fitness, t i is the true value of the i th sample, y i is the predicted value of the i th sample; and n is the number of training set samples.

[0108] Step S4, determination of the genetic operation:

[0109] Step S401, determination of the selection operation: the selection operations that can be selected include the roulette algorithm, the tournament algorithm, and the random traversal sampling algorithm.

[0110] In this step, the selection operation of the genetic operation adopts the roulette algorithm, and the formula is as follows:

[0111]

[0112] In the formula, F i is the fitness of each individual; N is the population size; k is the number of hidden layer neurons, and fi P is the product of k and the reciprocal of the fitness value; P i is the selection probability of each individual.

[0113] Step S402, determination of the crossover operation: the crossover operation of the genetic algorithm is performed between two individuals at random, and the selection of the crossover algorithm depends on the coding mode of the individual. In the crossover process, neither the good individual with high fitness nor the individual with good genes can be destroyed.

[0114] In this step, the real number fusion crossover method is selected, and the codes of two individuals at random positions that meet the crossover probability are randomly fused to generate two new individuals. The formula of the crossover operation is as follows:

[0115] a mk = a mk (1-b) + a nk b

[0116] a nk = a nk (1-b) + a mk b

[0117] In the formula, a mk is the mth individual a nk , a m is the nth individual a n , and b is a random number between 0 and 1.

[0118] Step S403, determination of the mutation operation: the mutation operation of the genetic algorithm mainly changes the value of the random coding position of a random individual. When the crossover operation approaches the optimal solution, the local random search capability of the mutation operation is used to accelerate the convergence to the optimal solution. At the same time, the mutation operation maintains the diversity of the population and avoids the phenomenon of premature convergence.

[0119] In this step, the real number mutation is adopted, and the formula is as follows:

[0120]

[0121] In the formula, a wk is the wth individual a w generated by mutation at a random position k; a max is the upper bound of the code; a min is the lower bound of the code; f(g) = b(1-num / maxGen) 2 ; b and r are random numbers between 0 and 1; num is the current iteration number; and maxGen is the preset maximum iteration number.

[0122] Step S404, determining other parameters: the number of iterations, population size, crossover probability and mutation probability need to be set before the genetic algorithm starts.

[0123] In this step, the number of iterations is set to 400, the population size is set to 500, the crossover probability is set to 0.7, and the mutation probability is set to 0.3.

[0124] Step S5, initial weight and threshold optimization:

[0125] The initial weights and thresholds of the network are optimized by the genetic algorithm. After the genetic algorithm calculates the global optimal solution, the weights and thresholds are assigned to the BP neural network, as shown in Figure 2 , the specific steps are as follows:

[0126] Step S501, population initialization: according to the structure of the BP neural network, the original population is randomly generated.

[0127] Step S502, chromosome coding: the chromosome is the weight and threshold of the BP neural network, and the chromosome is generated according to the order of weight first and threshold last.

[0128] Step S503, the training set A is taken as the model training data and substituted into the BP neural network with the initial weight and threshold, to obtain the error value of the BP neural network.

[0129] Step S504, calculate individual fitness: the error value obtained in step S503 is substituted into the fitness function to calculate the individual fitness and population average fitness of each individual.

[0130] Step S505, selection and replication: according to the order of the fitness, the individuals in the back are removed and the individuals in the front are copied for evolution.

[0131] Step S506, crossover and mutation: the population individuals are crossed and mutated according to the crossover and mutation formula.

[0132] Step S507, calculate the fitness value, and judge whether the individual after inheritance evolution is superior to the average fitness of the last generation. If the judgment meets the set requirements, continue to step S508 to complete decoding, if not, go to step S505 for the next cycle.

[0133] Step S508, decoding: the finally selected chromosome is decoded according to the order and coding method corresponding to step S502.

[0134] Step S509: assignment: as can be seen from step S508, the conversion of chromosome string to specific numerical value has been completed, and the set of weights and thresholds selected by the genetic algorithm is delivered to the BP neural network.

[0135] Step S6, test and optimization of BP neural network:

[0136] Step S601: input the test set B as an input parameter into the BP neural network with optimized weight value and threshold value.

[0137] Step S602, update the weight value and threshold value based on the error calculated by the BP neural network.

[0138] Step S603, calculate the error, and if the requirement is met, obtain the optimal weight value and threshold value of the BP neural network.

[0139] Step S7, collection of blasting index data of the mining area:

[0140] Step S701, rock mass density measurement: the volume method is used to weigh the mass of the test piece and measure the volume of the test piece, and then the density of the rock is calculated.

[0141] Step S702, measurement of uniaxial compressive strength and tensile strength of rock: drill core, cut test piece, polish and shape, etc. are performed on the site rock mass, and the measurement of uniaxial compressive strength and tensile strength of rock is completed.

[0142] In this step, the Brazilian split test can be used for testing.

[0143] Step S703, rock mass integrity coefficient K v Calculation: the rock mass integrity coefficient is the square of the ratio of the longitudinal wave velocity of the rock mass to the longitudinal wave velocity of the rock, as shown in the following formula:

[0144]

[0145] In the formula, V pm is the longitudinal wave velocity of the rock mass, and V pr is the longitudinal wave velocity of the rock. The average value of the longitudinal wave velocity of the rock mass is obtained by measuring multiple points in multiple working faces such as tunneling roadway, and the rock in the working face is processed into a standard sample for indoor rock longitudinal wave velocity measurement.

[0146] In this step, the ZBL-U5200 non-metal ultrasonic detector can be used for acoustic detection. The instrument records the time taken by the acoustic wave to pass through the rock mass or rock, and then measures the longitudinal wave velocity of the rock mass or rock by dividing the length by the time.

[0147] Step S704, average block size and large block rate calculation: single-hole blasting test is performed on the rock mass of the mining area, the rock blocks are cleaned and weighed respectively, and the average block size and large block rate are calculated.

[0148] Step S8, intelligent evaluation of rock mass explosibility zoning of the mining area:

[0149] The six rock indexes obtained by the test are input into the BP neural network, and five values corresponding to the five blastability partitions of the rock mass in the mine are output. The value at the corresponding blastability partition level is 1, and the value at other blastability partition levels is 0.

[0150] In this step, if the output values are I: 0.0000, II: 1.0000, III: 0.0000, IV: 0.0000, and V: 0.0000, the mine rock blastability partition is II, and the blastability description is easy to explode.

Claims

1. A neural network-based intelligent evaluation method for the blastability zoning of rock masses in mining areas, characterized in that... The method includes the following steps: Step S1: Generating the dataset: Step S101: Dataset generation: Based on the summary of existing literature, the standard for rock mass blastability zoning is summarized, and the zoning interval values ​​of each individual evaluation factor corresponding to each rock mass blastability level are determined. Using the ROUND and RAND functions in Excel, the input layer corresponding to each rock mass blastability level is randomly selected within the index limit range to generate sample data. Step S102, Dataset Standardization: Normalize the dataset to obtain normalized data; Step S103: Dataset partitioning: Divide the dataset into a training set and a test set; Step S2: Establishing the BP neural network Six indicators were selected as the input layer of the BP neural network: rock density, rock uniaxial compressive strength, rock uniaxial tensile strength, rock mass integrity coefficient, average block size, and large block ratio. The rock mass blastability was divided into five levels: extremely blastable, blastable, moderate, difficult to blast, and extremely difficult to blast, which were used as the output layer of the BP neural network. Step S3, setting up the genetic algorithm: Step S301: Determine the encoding range and length: The weights and thresholds of the BP neural network are in the range of [-1, 1]. The encoding range of the chromosomes in the genetic algorithm corresponds to the weights and thresholds of the BP neural network. The formula for calculating the encoding length is as follows: l = e·k + k·u + (e + u) In the formula, l is the encoding length, k is the number of hidden layer neurons, e is the number of input layers, and u is the number of output layers; Step S302, Determining the encoding method: Use the real number encoding method, and the encoded value is a random number within the encoding range; Step S303: Determine the fitness function: Assign the weight and threshold information of the individual to the BP neural network, and train the BP neural network using the training dataset. The fitness calculation formula is shown below: In the formula, F is the fitness, and t i Let y be the true value of the i-th sample. i Let be the predicted value of the i-th sample; n is the number of samples in the training set. Step S4, Determining the genetic operation: Step S401, Determining the selection operation: The selection operation of the genetic operation adopts the roulette wheel algorithm, and the formula is: In the formula, F i The fitness of each individual, N is the population size, k is the number of hidden layer neurons, and f i P is the product of k and the reciprocal of fitness. i The probability of choice for each individual; Step S402, Determining the crossover operation: Select the real number fusion crossover method, randomly merge the codes of two individuals that meet the crossover probability at random positions to generate two new individuals. The crossover operation formula is shown below: a mk =a m (1-b)+a n b a nk =a n (1-b)+a m b In the formula, a mk With a nk For the m-th individual a m With the nth individual a n Individuals generated by cross-fusion at random position k; b is a random number between [0,1]. Step S403, Determining the mutation operation: Real number mutation is used, and the formula is as follows: In the formula, a wk For the w-th individual a w The individual generated by mutation at random position k; a max a is the upper bound of the encoding; min This is the lower bound of the encoding; f(g) = b(1 - num / maxGen) 2 b and r are random numbers between [0,1]. num is the current iteration number; maxGen is the preset maximum iteration number; Step S404: Determine other parameters: Before starting the genetic algorithm, it is necessary to set the number of iterations, population size, crossover probability, and mutation probability; Step S5, Initial weights and threshold optimization: The initial weights and thresholds of the network are optimized by a genetic algorithm. After the genetic algorithm calculates the global optimal solution, the weights and thresholds are assigned to the BP neural network. Step S6: Testing and tuning the BP neural network: Step S601: Use the test set B as input parameters and input it into the BP neural network with optimized weights and thresholds; Step S602: Update the weights and thresholds based on the error calculated by the BP neural network; Step S603: Calculate the error. If the requirements are met, obtain the optimal weights and thresholds of the BP neural network. Step S7: Collection of blasting index data in the mining area: Step S701: Measure the density of the rock mass; Step S702: Measure the uniaxial compressive strength and tensile strength of the rock; Step S703: Calculate the rock mass integrity factor K v : In the formula V pm V represents the longitudinal wave velocity of the rock mass. pr The longitudinal wave velocity of the rock; Step S704: Calculate the average block size and the percentage of large blocks; Step S8, Intelligent evaluation of the blastability of rock mass in the mining area: The six rock indices obtained from the experiment were input into a BP neural network, and five output values ​​were generated corresponding to five blastability zones of the rock mass in the mining area. The value at the corresponding blastability zone level was 1, and the value at other blastability zone levels was 0.

2. The intelligent evaluation method for the blastability zoning of rock masses in mining areas based on neural networks according to claim 1, characterized in that... In step S102, deviation standardization is used to process the original data. The standardization formula is shown below: In the formula, The maximum value of the j-th indicator; The minimum value of the j-th indicator; a ij b is the i-th element of the current indicator; ij The value is a standardized value, ranging from [0,1].

3. The intelligent evaluation method for the blastability zoning of rock masses in mining areas based on neural networks according to claim 1, characterized in that... The specific steps for establishing step S2 are as follows: Step S201: Determining the number of neurons in the hidden layer: The number of hidden layer neurons is determined using the following formula: Where k is the number of hidden layer neurons; e is the number of input layer neurons; u is the number of output layer neurons; and a is a constant between 1 and 10. Step S202: Determining the activation function and training function of the BP neural network: Select the hidden layer activation function, output layer activation function, and training function as needed; Step S203: Determining other parameters of the BP neural network: Select the maximum number of learning iterations, initial learning rate, initial momentum factor, and required training accuracy for the BP neural network as needed.

4. The intelligent evaluation method for the blastability zoning of rock masses in mining areas based on neural networks according to claim 1, characterized in that... The specific steps of step S5 are as follows: Step S501, Population Initialization: Based on the structure of the BP neural network, randomly generate the original population; Step S502, Chromosome Encoding: Chromosomes are the weights and thresholds of a BP neural network, and are generated by arranging them in a structural order of weights first and thresholds last. Step S503: Substitute the training set A as the model training data into the BP neural network that initializes the weights and thresholds to obtain the error value of the BP neural network. Step S504: Calculate individual fitness: Substitute the error value obtained in step S503 into the fitness function to calculate the individual fitness of each individual and the population average fitness. Step S505: Selecting for replication: Arrange individuals in order of fitness. In this evolutionary process, according to the roulette wheel algorithm, individuals ranked lower are removed, and individuals ranked higher are replicated and evolved. Step S506, Crossover and Mutation: Perform crossover and mutation on individuals in the population according to the crossover and mutation formula; Step S507: Calculate the fitness value and determine whether the individual that has completed the inheritance evolution is superior to the average fitness of the previous generation. If the determination meets the requirements, continue to step S508 to complete the decoding. If not, go to step S505 to start the next loop. Step S508, Decoding: Decode the winning chromosomes according to the sorting and encoding method corresponding to step S502; Step S509: Assignment: As can be seen from step S508, the conversion from chromosome string to specific numerical value has been completed. The set of weights and thresholds selected by the genetic algorithm is delivered to the BP neural network.

Images