A method and device for predicting blasting fragmentation based on neural network optimization algorithm

Abstract

The present invention discloses a method and device for predicting blasting fragmentation based on a neural network optimization algorithm. The present invention adopts a random forest and a particle swarm optimization algorithm to respectively perform feature selection processing on blasting fragmentation influencing data, and uses the indicators selected by different algorithms to comprehensively determine the optimal blasting fragmentation influencing indicator; then, based on the optimal blasting fragmentation influencing indicator, a model is trained to obtain a blasting fragmentation prediction model; finally, the blasting fragmentation prediction model is used to obtain a blasting fragmentation prediction result of a mine; thus, the present invention comprehensively determines the indicator data most relevant to the blasting fragmentation based on the features selected by the random forest and particle swarm optimization algorithms, realizes feature selection based on data-driven, thereby avoiding the problem of being easily affected by human subjective experience in traditional technologies, based on which the accuracy and objectivity of feature selection are improved, thereby increasing the accuracy of subsequent blasting fragmentation prediction.

Description

Technical Field

[0001] The invention belongs to the technical field of blasting engineering, and in particular relates to a method and a device for predicting blasting fragmentation based on a neural network optimization algorithm. Background Art

[0002] Blasting is a commonly used mining and construction engineering technology, which is widely used in coal, metal mines, highways, hydropower projects, tunnels and other fields. During the blasting operation, the blasting fragmentation is an important indicator to measure the blasting effect. The distribution and gradation of the fragmentation have an important impact on the subsequent excavation, loading, transportation and other work; therefore, accurate prediction of blasting fragmentation is of great significance for optimizing blasting plans, improving work efficiency and reducing costs.

[0003] At present, in the early stage of application, people mainly rely on empirical formula summary, rock mechanics model calculation and other methods to predict blasting fragmentation. Among them, the above methods have the disadvantages of insufficient accuracy and strong subjectivity. Recently, with the rapid development of machine learning algorithms, machine learning algorithms have also been gradually applied to fragmentation prediction. Among them, machine learning requires a large amount of sample data for model training, and in actual applications, most of them are based on expert experience to select fixed indicator data for prediction. It has the following shortcomings: indicator selection is easily affected by human subjective experience, resulting in insufficient stability and poor generalization ability, which will affect the accuracy of subsequent fragmentation prediction. Therefore, based on the above shortcomings, how to provide a prediction method that can objectively and accurately select blasting fragmentation influencing indicators and predict blasting fragmentation based on this has become a problem to be solved urgently. Summary of the invention

[0004] The purpose of the present invention is to provide a method and device for predicting blasting fragmentation based on a neural network optimization algorithm, so as to solve the problem that the accuracy of subsequent fragmentation prediction will be affected by the prior art using expert experience to select blasting fragmentation influencing indicators.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, a method for predicting blasting fragmentation based on a neural network optimization algorithm is provided, comprising:

[0007] Obtain the blasting block size impact data corresponding to the specified mine;

[0008] Using a random forest algorithm, feature selection processing is performed on the blasting block size influence data to screen out a first blasting block size influence index from the blasting block size influence data;

[0009] Based on the blasting block size influence data, a particle population is constructed, wherein the particle population includes a number of individual particles, each element in the position vector of any individual particle corresponds to an index data in the blasting block size influence data, the position vector of any individual particle is a binary vector, and any element in the position vector corresponding to any individual particle is 1, indicating that the index data corresponding to the any element is selected as an initial second blasting block size influence index, and any element in the position vector corresponding to any individual particle is 0, indicating that the index data corresponding to the any element is not selected;

[0010] Using an improved particle swarm optimization algorithm, the particle population is optimized to obtain an optimal individual particle, and using the optimal individual particle to determine a second blasting block size influence index from the blasting block size influence data;

[0011] Determine the optimal blasting fragmentation influence index according to the first blasting fragmentation influence index and the second blasting fragmentation influence index;

[0012] Based on the optimal blasting block size influence index, obtaining historical blasting data;

[0013] A neural network optimization algorithm is used, and the historical blasting block size influence data in the historical blasting data is used as input, and the blasting block size prediction value of the historical mine is used as output to train the neural network model, so as to obtain a blasting block size prediction model after the training is completed;

[0014] The blasting fragmentation prediction model is used to obtain the blasting fragmentation prediction result of the designated mine.

[0015] Based on the above disclosed content, after obtaining the blasting block size influence data of the specified mine, the present invention first uses the random forest algorithm to analyze the correlation between each indicator in the blasting block size influence data and the blasting block size, and uses this to select the feature importance, so as to screen out the first blasting block size influence indicator from the blasting block size influence data; at the same time, the present invention also constructs a particle population based on the blasting block size influence data, wherein the position vector of any individual particle in the particle population is a binary vector, and the value of each internal element is used to determine whether the indicator data corresponding to the element is selected; based on this, the improved The particle swarm optimization algorithm is used to optimize the particle population to obtain the optimal individual particle; then, according to the position vector of the optimal individual particle, the second blasting block size influence index can be determined from the blasting block size influence data; then, according to the aforementioned first and second blasting block size influence indexes, the optimal blasting block size influence index can be determined; after obtaining the optimal blasting block size influence index, the optimal blasting block size influence index can be used, and the neural network optimization algorithm can be used to train the neural network model; finally, based on the trained model, the blasting block size prediction result of the specified mine can be obtained.

[0016] Through the above design, the present invention adopts random forest and particle swarm optimization algorithms to respectively perform feature selection processing on the blasting block size influencing data, and uses the indicators selected by different algorithms to comprehensively determine the optimal blasting block size influencing indicator; then, based on the optimal blasting block size influencing indicator, the model is trained to obtain the blasting block size prediction model; finally, the blasting block size prediction model is used to obtain the blasting block size prediction result of the specified mine; thus, the present invention comprehensively determines the indicator data most relevant to the blasting block size based on the features selected by random forest and particle swarm optimization algorithms, thus realizing data-driven feature selection, thereby avoiding the problem of traditional technology being easily affected by human subjective experience, based on this, improving the accuracy and objectivity of feature selection, thereby increasing the accuracy of subsequent blasting block size prediction, and therefore, is very suitable for large-scale application and promotion.

[0017] In a possible design, based on the blasting blockiness impact data, a particle population is constructed, including:

[0018] According to the blasting block size influence data, a plurality of initial particles are constructed, wherein each element in the position vector of any initial particle corresponds to an indicator data in the blasting block size influence data;

[0019] Performing initialization processing on each initial particle, and after the initialization processing, performing position mapping processing on each initial particle, so as to obtain individual particles corresponding to each initial particle after the position mapping processing;

[0020] Using individual particles corresponding to each initial particle to form the particle population;

[0021] Among them, using the improved particle swarm optimization algorithm to optimize the particle population to obtain the optimal individual particle includes:

[0022] Obtaining a fitness function, wherein the fitness function is used to characterize the selection accuracy of the initial second blasting block size influence index selected based on any individual particle, and the greater the fitness of any individual particle, the higher the selection accuracy of the initial second blasting block size influence index selected based on the any individual particle;

[0023] Initialize the optimization number t to 1, and calculate the fitness of each individual particle in the particle population at the tth optimization;

[0024] Based on the calculated fitnesses, the global optimal fitness is determined;

[0025] Determine whether the global optimal fitness is greater than or equal to the first fitness threshold, or whether t reaches the maximum number of optimization times;

[0026] If not, a position update weight is calculated, and based on the position update weight, the position vector of each individual particle is updated to obtain a number of updated individual particles after the position vector is updated;

[0027] Performing position mapping processing on each updated individual particle to obtain a number of mapped individual particles, and using the number of mapped individual particles to form a new particle population;

[0028] The particle population is updated to the new particle population, t is incremented by 1, and the fitness of each individual particle in the particle population at the t-th optimization search is recalculated until the global optimal fitness is greater than or equal to the first fitness threshold, or t reaches the maximum number of optimization searches, and the individual particle corresponding to the global optimal fitness is used as the optimal individual particle;

[0029] Correspondingly, using the optimal individual particles, determining a second blasting fragmentation influence index from the blasting fragmentation influence data includes:

[0030] The initial second blasting fragmentation influence index corresponding to the element with a value of 1 in the position vector corresponding to the optimal individual particle is used as the second blasting fragmentation influence index.

[0031] In a possible design, position mapping is performed on each initial particle to obtain individual particles corresponding to each initial particle after the position mapping, including:

[0032] For any initial particle, the mapping value of each element in the position vector corresponding to the initial particle is calculated according to the following formula (1);

[0033]

[0034] In the above formula (1), f(r) represents the mapping value of the rth element in the position vector corresponding to any initial particle, x r represents the rth element in the position vector corresponding to any of the initial particles, wherein r=1, 2, ..., R, and R represents the total number of elements in the position vector corresponding to any of the initial particles;

[0035] Generate a first random number for any of the initial particles, wherein the value interval of the first random number is [0, 1];

[0036] Based on the first random number and the mapping value of each element in the position vector corresponding to any initial particle, the position vector of any initial particle is mapped to obtain the individual particle corresponding to any initial particle after the mapping process, wherein, when performing the mapping process, for any element in the position vector corresponding to any initial particle, if the first random number is greater than or equal to the mapping value of any element, the any element is set to 1; if the first random number is less than the mapping value of any element, the any element is set to 0.

[0037] In one possible design, the fitness function is:

[0038]

[0039] In formula (2), F represents the fitness function, κ represents the selection accuracy, Z represents the number of initial second blasting block size influence indicators determined based on individual particles, N represents the total number of indicators in the blasting block size influence data, η1, η2 both represent selection weights, wherein the selection accuracy is obtained by using a specified classification algorithm to classify the initial second blasting block size influence indicators determined based on individual particles;

[0040] The calculation of the position update weight includes:

[0041] The position update weight is calculated using the following formula (3);

[0042]

[0043] In the above formula (3), ω represents the position update weight, ω max represents the maximum weight, ω min represents the minimum weight, and T represents the maximum number of optimization times.

[0044] In a possible design, a neural network optimization algorithm is used, and the historical blasting block size impact data in the historical blasting data is used as input, and the blasting block size prediction value of the historical mine is used as output to train the neural network model, including:

[0045] Obtaining model parameters of the neural network model, and constructing a gray wolf population based on the model parameters, wherein the gray wolf population includes a plurality of gray wolves, and the position of any gray wolf is used to represent a set of model parameters;

[0046] Initialize the number of iterations s to 1, and calculate the fitness of each gray wolf in the gray wolf population at the sth iteration, wherein the fitness of any gray wolf is calculated based on the label value and output data corresponding to the input data of the target neural network corresponding to the any gray wolf, the input data of the target neural network is the historical blasting data, and the model parameters of the target neural network are the model parameters corresponding to the position of any gray wolf;

[0047] According to the order of fitness from large to small, each gray wolf is sorted, and the first three gray wolves are regarded as α wolf, β wolf and δ wolf in turn, and the gray wolves ranked after the third place are regarded as follower wolves;

[0048] Determine whether the fitness of wolf α is greater than or equal to the second fitness threshold, or whether s reaches the maximum number of iterations;

[0049] If not, calculate the gray wolf update factor and the distance between each follower wolf and α wolf, β wolf and δ wolf;

[0050] Update the positions of the α wolf, β wolf and δ wolf, and update the positions of the following wolves according to the distances between the following wolves and the α wolf, β wolf and δ wolf, and the gray wolf update factor;

[0051] Use the updated positions of α wolf, β wolf, δ wolf and each follower wolf to form a new gray wolf population;

[0052] Increment s by 1 and update the gray wolf population to the new gray wolf population, and recalculate the fitness of each gray wolf in the gray wolf population at the sth iteration, until the fitness of wolf α is greater than or equal to the second fitness threshold, or s reaches the maximum number of iterations, and the model parameters represented by the position of wolf α are used as the optimal model parameters;

[0053] Based on the optimal model parameters, the neural network model is optimized to obtain an optimized neural network model, and the blasting fragmentation prediction model is obtained using the optimized neural network model.

[0054] In one possible design, the gray wolf update factor is calculated to include:

[0055] The gray wolf renewal factor is calculated using the following formula (4);

[0056]

[0057] In the above formula (4), θ represents the gray wolf update factor, and S represents the maximum number of iterations;

[0058] Accordingly, updating the position of each follower wolf according to the distance between each follower wolf and the α wolf, the β wolf and the δ wolf, and the gray wolf update factor, includes:

[0059] For any follower wolf, based on the gray wolf update factor, the first coefficient is calculated using the following formula (5);

[0060] A=2×θ×b1-θ (5)

[0061] In the above formula (5), A represents the first coefficient, and b1 represents a second random number between [0, 1];

[0062] Based on the first coefficient and the distance between any of the following wolves and the α wolf, the β wolf and the δ wolf, respectively calculating the tendency amount of any of the following wolves to move toward the α wolf, the β wolf and the δ wolf;

[0063] According to the trend amount of any follower wolf moving toward the α wolf, the β wolf, and the δ wolf, a moving trend weight between any follower wolf and the α wolf, the β wolf, and the δ wolf is calculated;

[0064] The position of any following wolf is updated by using the movement trend weights between any following wolf and the α wolf, the β wolf and the δ wolf, so as to obtain the updated any following wolf after the position is updated.

[0065] In a possible design, according to the trend amount of any follower wolf moving toward α wolf, β wolf and δ wolf, the moving trend weight between any follower wolf and α wolf, β wolf and δ wolf is calculated, including:

[0066] The following formulas (6), (7) and (8) are used in sequence to calculate the movement trend weights between any following wolf and the α wolf, the β wolf and the δ wolf;

[0067]

[0068] In the above formula (6), W α,j represents the moving trend weight between any following wolf and α wolf, X α,j ,X β,j ,X δ,j represents the trend amount of any following wolf moving toward wolf α, the trend amount of moving toward wolf β, and the trend amount of moving toward wolf δ, respectively. represents the learning factor;

[0069] In the above formula (7), W β,j represents the moving trend weight between any following wolf and β wolf;

[0070] In the above formula (8), W δ,j represents the moving trend weight between any following wolf and δ wolf;

[0071] Accordingly, using the movement trend weights between any of the following wolves and the α wolf, the β wolf, and the δ wolf to update the position of any of the following wolves includes:

[0072] The following formula (9) is used to update the position of any following wolf;

[0073]

[0074] In the above formula (9), represents the updated position of any following wolf, A vector representation of the tendency of any of the following wolves to move toward the alpha wolf.

[0075] In a second aspect, a blasting fragmentation prediction device based on a neural network optimization algorithm is provided, comprising:

[0076] An acquisition unit, used for acquiring blasting block size impact data corresponding to a specified mine;

[0077] A feature selection unit, configured to perform feature selection processing on the blasting block size influence data using a random forest algorithm, so as to screen out a first blasting block size influence index from the blasting block size influence data;

[0078] A feature selection unit is used to construct a particle population based on the blasting block size influence data, wherein the particle population includes a plurality of individual particles, each element in the position vector of any individual particle corresponds to an index data in the blasting block size influence data, the position vector of any individual particle is a binary vector, and any element in the position vector corresponding to any individual particle is 1, indicating that the index data corresponding to the any element is selected as an initial second blasting block size influence index, and any element in the position vector corresponding to any individual particle is 0, indicating that the index data corresponding to the any element is not selected;

[0079] A feature selection unit is used to use an improved particle swarm optimization algorithm to perform optimization processing on the particle population to obtain optimal individual particles, and use the optimal individual particles to determine a second blasting block size influence index from the blasting block size influence data;

[0080] The feature selection unit is further used to determine the optimal blasting fragmentation influence index according to the first blasting fragmentation influence index and the second blasting fragmentation influence index;

[0081] A training unit, used for acquiring historical blasting data based on the optimal blasting block size influence index;

[0082] The training unit is further used to adopt a neural network optimization algorithm, and use the historical blasting block size impact data in the historical blasting data as input and the blasting block size prediction value of the historical mine as output to train the neural network model, so as to obtain a blasting block size prediction model after the training is completed;

[0083] The fragmentation prediction unit is used to obtain the blasting fragmentation prediction result of the specified mine by using the blasting fragmentation prediction model.

[0084] In a third aspect, another device for predicting blasting fragmentation based on a neural network optimization algorithm is provided. Taking the device as an electronic device as an example, it includes a memory, a processor and a transceiver that are communicatively connected in sequence, wherein the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute the method for predicting blasting fragmentation based on a neural network optimization algorithm as in the first aspect or any possible design of the first aspect.

[0085] In a fourth aspect, a storage medium is provided, on which instructions are stored. When the instructions are executed on a computer, the blasting fragmentation prediction method based on a neural network optimization algorithm as in the first aspect or any possible design of the first aspect is executed.

[0086] In a fifth aspect, a computer program product comprising instructions is provided, which, when executed on a computer, causes the computer to execute the blasting fragmentation prediction method based on a neural network optimization algorithm as described in the first aspect or any possible design of the first aspect.

[0087] Beneficial effects:

[0088] (1) The present invention comprehensively determines the index data most relevant to the blasting fragmentation based on the features selected by the random forest and particle swarm optimization algorithms. In this way, data-driven feature selection is achieved, thereby avoiding the problem of traditional technologies being easily affected by human subjective experience. Based on this, the accuracy and objectivity of feature selection are improved, thereby increasing the accuracy of subsequent blasting fragmentation prediction. Therefore, the present invention is very suitable for large-scale application and promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 A schematic diagram of the steps of a blasting fragmentation prediction method based on a neural network optimization algorithm provided in an embodiment of the present invention;

[0090] Figure 2 A schematic diagram of the structure of a blasting fragmentation prediction device based on a neural network optimization algorithm provided in an embodiment of the present invention;

[0091] Figure 3 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0092] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0093] It should be understood that although the terms first, second, etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another unit. For example, a first unit can be referred to as a second unit, and similarly, a second unit can be referred to as a first unit without departing from the scope of the exemplary embodiments of the present invention.

[0094] It should be understood that the term "and / or" that may appear in this article is merely a description of the association relationship between associated objects, indicating that three relationships may exist. For example, A and / or B can represent three situations: A exists alone, B exists alone, and A and B exist at the same time. The term " / and" that may appear in this article describes another type of association object relationship, indicating that two relationships may exist. For example, A / and B can represent two situations: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this article generally indicates that the previous and next associated objects are in an "or" relationship.

[0095] Example:

[0096] See also Figure 1As shown, the blasting fragmentation prediction method based on the neural network optimization algorithm provided in this embodiment comprehensively determines the index data most relevant to the blasting fragmentation based on the features selected by the random forest and particle swarm optimization algorithms, and then, based on the determined most relevant index data, performs model training to obtain a blasting fragmentation prediction model; finally, based on the blasting fragmentation prediction model, a blasting fragmentation prediction result of a specified mine can be obtained; in this way, the method realizes feature selection based on data-driven, thereby avoiding the problem of being easily affected by human subjective experience in traditional technologies, based on which, the accuracy and objectivity of feature selection are improved, thereby increasing the accuracy of subsequent blasting fragmentation prediction, and therefore, is very suitable for large-scale application and promotion; wherein, the method can be but not limited to running on the blasting fragmentation prediction end side, optionally, the blasting fragmentation prediction end can be but not limited to a personal computer (PC), a tablet computer or a smart phone, it can be understood that the aforementioned execution subject does not constitute a limitation on the embodiments of the present application, and accordingly, the operation steps of the method can be but not limited to the following steps S1 to S8.

[0097] S1. Obtain the blasting block size influence data corresponding to the designated mine; in this embodiment, the blasting block size influence data may include but is not limited to blasting parameters, rock structure parameters of the designated mine and rock mechanical property parameters of the designated mine; wherein, the blasting parameters may include but are not limited to the first row of hole resistance line, chassis resistance line, blasting hole spacing, borehole diameter, minimum resistance line, explosive consumption, blast hole plugging length and single hole charge, etc. Similarly, the rock structure parameters may include but are not limited to rock size, rock bulk density, cohesion and internal friction angle, etc., while the rock mechanical parameters may include but are not limited to elastic modulus, compressive strength, tensile strength and Poisson's ratio, etc.; of course, the aforementioned blasting block size influence data may be pre-stored in the blasting block size prediction terminal, and each data in the blasting block size influence data is used as an indicator; in this way, the present embodiment needs to first select the indicator most relevant to the blasting block size from the aforementioned multiple indicators, so as to conduct model training based on the selected indicator later.

[0098] Among them, this embodiment adopts two methods to select indicators, and combines the indicators selected by the two methods to obtain the optimal blasting fragmentation influence indicator; optionally, the indicator selection process can be but is not limited to the following steps S2 to S5.

[0099] S2. Use the random forest algorithm to perform feature selection processing on the burst block size influencing data to screen out the first burst block size influencing index from the burst block size influencing data; in this embodiment, it is equivalent to analyzing the burst block size influencing factors through the random forest algorithm, so as to perform feature importance screening, so that the indicators ranked in the top 3 or 5 in importance can be used as the first burst block size influencing index; in this embodiment, there are two main methods for random forest to measure feature importance, namely, mean impurity reduction and Gini importance, among which, mean impurity reduction (MeanDecreaseImpurity) is a method based on impurity reduction, that is, in the process of constructing each decision tree, the number of times each feature is used in the decision tree is calculated, and the impurity reduction caused by each feature after dividing the node is calculated; finally, the impurity reduction of each feature in all decision trees is averaged to obtain the feature importance evaluation value; and Gini importance (Gini The feature importance evaluation method based on the random forest algorithm is a method based on the Gini coefficient. The Gini coefficient measures the probability of a feature being misclassified when a sample is randomly selected. In the process of constructing a random forest, the average Gini coefficient of each feature in all decision trees is calculated to obtain the feature importance evaluation value. Of course, the aforementioned feature importance evaluation based on the random forest algorithm is a common feature selection technique, and its principle will not be repeated here.

[0100] In this way, after determining the first blasting fragmentation influence index from the blasting fragmentation influence data based on the aforementioned step S2, another feature selection method can be used to perform feature selection (i.e., index selection) so that the optimal blasting fragmentation influence index can be obtained based on the indicators determined by the two methods; wherein the second feature selection method can be but is not limited to the following steps S3 and S4.

[0101] S3. Based on the blasting block size influence data, a particle population is constructed, wherein the particle population includes a number of individual particles, and each element in the position vector of any individual particle corresponds to an indicator data in the blasting block size influence data. The position vector of any individual particle is a binary vector, and any element in the position vector corresponding to any individual particle is 1, indicating that the indicator data corresponding to the any element is selected as an initial second blasting block size influence indicator, and any element in the position vector corresponding to any individual particle is 0, indicating that the indicator data corresponding to the any element is not selected.

[0102] In the specific implementation, it is equivalent to generating a number of particles, each particle contains all the indicators in the blasting block size impact data, and then, an initial position is assigned to each particle; then, the optimal particle is obtained by optimizing the position of each particle; finally, the second blasting block size impact indicator can be determined based on the optimal particle; optionally, the construction process of the aforementioned particle population can be but is not limited to the following steps S31 to S33.

[0103] S31. According to the blasting block size influence data, a number of initial particles are constructed, wherein each element in the position vector of any initial particle corresponds to an index data in the blasting block size influence data; in the present embodiment, taking the aforementioned blasting block size influence data as an example, it contains 16 indexes, then the position vector of each initial particle contains 16 elements, each element corresponds to an index, such as element 1 corresponds to the index data of the first row of holes resistance line, element 2 corresponds to the index data of the chassis resistance line, etc., in this way, each particle represents a set of full index data; based on this, by assigning different initial values ​​to the position vector of each particle, each initial particle can have a different position vector; finally, the optimal position is optimized by using the optimization algorithm to obtain the optimal particle; at the same time, in order to make the position vector of the initial particle associated with the feature selection, it is also necessary to map the position vector so that the elements inside it are 1 or 0, so as to select features based on the element values; optionally, the initialization processing and position mapping processing of each initial particle can be but not limited to the following step S32.

[0104] S32. Initialize each initial particle, and after the initialization process, perform position mapping on each initial particle to obtain individual particles corresponding to each initial particle after the position mapping process; in the specific implementation, since the position mapping process of each initial particle is the same, the following will take any initial particle as an example for explanation, and its position mapping process can be but is not limited to the following steps S32a~S32c.

[0105] S32a. For any initial particle, the mapping value of each element in the position vector corresponding to the initial particle is calculated according to the following formula (1).

[0106]

[0107] In the above formula (1), f(r) represents the mapping value of the rth element in the position vector corresponding to any initial particle, x r Represents the rth element in the position vector corresponding to any of the initial particles, where r=1, 2, ..., R, and R represents the total number of elements in the position vector corresponding to any of the initial particles.

[0108] In the present embodiment, it has been explained above that the position vector of each initial particle will be initialized, that is, an initial value will be assigned. Thus, by substituting each element in the position vector of any of the aforementioned initial particles into the aforementioned formula (1), each element can be mapped to a value between 0 and 1, thereby obtaining a mapping value corresponding to each element. After obtaining the mapping value of each element, the position mapping of any of the initial particles can be completed based on each mapping value, and the process can be, but is not limited to, as shown in the following steps S32b and S32c.

[0109] S32b. Generate a first random number for any of the initial particles, wherein the value interval of the first random number is [0,1]. In the present embodiment, the first random number is compared with the mapping value of each element in the corresponding position vector of any of the initial particles, thereby encoding each element as 0 or 1, and obtaining the corresponding individual particle after the encoding is completed. The aforementioned process may be, but is not limited to, as shown in the following step S32c.

[0110] S32c. Based on the first random number and the mapping value of each element in the position vector corresponding to any initial particle, the position vector of any initial particle is mapped to obtain the individual particle corresponding to any initial particle after the mapping process, wherein, when performing the mapping process, for any element in the position vector corresponding to any initial particle, if the first random number is greater than or equal to the mapping value of any element, the any element is set to 1; if the first random number is less than the mapping value of any element, the any element is set to 0.

[0111] In this embodiment, an example is used to illustrate the above step S32c. It is assumed that the position vector of any initial particle contains 6 elements (for the convenience of example, the above 16 indicators are not taken as an example), wherein the mapping values ​​corresponding to the 6 elements are 0.62, 0.52, 0.8, 0.9, 0.4 and 0.85 respectively, and the above first random number is 0.55. Then, after the position vector of any initial particle is mapped, it becomes {1,0,1,1,0,1}. Thus, the position vector of the individual particle corresponding to any initial particle is {1,0,1,1,0,1}. Based on this, the indicator can be selected through the position vector, that is, the indicators corresponding to element 1, element 3, element 4 and element 6 are selected as an initial second blasting block size influencing indicator, while the indicators corresponding to element 2 and element 5 are not selected. Of course, the above example is only an example. When the mapping values ​​are different, the mapping process of the position vector is the same as the above example, which will not be repeated here.

[0112] Thus, through the aforementioned steps S32a to S32c, the mapping process of each initial particle can be completed to obtain the corresponding individual particles; then, using the aforementioned individual particles, a particle population can be formed, and the process can be but not limited to as shown in the following step S33.

[0113] S33. Utilize the individual particles corresponding to each initial particle to form the particle population.

[0114] Based on the aforementioned steps S31 to S33, a particle population in which each particle represents a set of index data can be constructed based on the blasting block size influence data; in this way, the optimal individual particle is obtained by optimizing the particle population; and then, using the position vector of the optimal individual particle, the second blasting block size influence index can be determined from the blasting block size influence data; wherein, the optimization process of the particle population can be but is not limited to as shown in the following step S4.

[0115] S4. Use the improved particle swarm optimization algorithm to optimize the particle population to obtain the optimal individual particles, and use the optimal individual particles to determine the second blasting block size influence index from the blasting block size influence data; in the specific implementation, for example, but not limited to, the following steps S41 to S47 can be used to complete the optimization of the particle population.

[0116] S41. Obtain a fitness function, wherein the fitness function is used to characterize the selection accuracy of the initial second blasting block size influence index selected based on any individual particle, and the greater the fitness of any individual particle, the higher the selection accuracy of the initial second blasting block size influence index selected based on the any individual particle; in specific applications, this embodiment aims to achieve the highest accuracy of the selected index, then, its fitness function is the selection accuracy of the index; therefore, a fitness function can be designed to minimize the number of feature selections while maximizing the selection accuracy, wherein an example fitness function can be, but is not limited to, as shown in the following formula (2).

[0117]

[0118] In the above formula (2), F represents the fitness function, κ represents the selection accuracy, Z represents the number of initial second blasting block size influence indicators determined based on individual particles, N represents the total number of indicators in the blasting block size influence data, η1, η2 both represent selection weights, wherein the selection accuracy is obtained by using a specified classification algorithm to classify the initial second blasting block size influence indicators determined based on individual particles; in this embodiment, for example, but not limited to, the C4.5 decision tree algorithm can be used to evaluate the initial second blasting block size influence indicators selected based on the position vector of each individual particle, so as to obtain the corresponding selection accuracy; based on this, each individual particle can use its corresponding position vector to select its own initial second blasting block size influence indicator after each position update; then, the C4.5 decision tree algorithm is used to evaluate the accuracy of the selected indicator; finally, the evaluated accuracy is substituted into the above formula (2), and the fitness of each individual particle can be obtained; of course, the C4.5 decision tree algorithm is a commonly used classification algorithm, and its principle will not be repeated.

[0119] Based on this, the aforementioned fitness function is designed to reduce the number of selected indicators while ensuring the accuracy of the selected indicators; and after constructing the aforementioned fitness function, particle optimization processing can be performed, and the process can be but is not limited to the following steps S42 to S47.

[0120] S42. Initialize the number of optimization searches t to 1, and calculate the fitness of each individual particle in the particle population at the tth optimization search; in this embodiment, the position vector of each individual particle at the tth optimization search (when t is 1, it is the initial position vector) is used to determine the initial second blasting block size influence index corresponding to each individual particle; then, the C4.5 decision tree algorithm is used to evaluate the selection accuracy of the initial second blasting block size influence index corresponding to each individual particle; finally, the selection accuracy is substituted into the aforementioned formula (2), and the fitness of each individual particle at the tth optimization search can be obtained; of course, the aforementioned C4.5 decision tree algorithm is a commonly used technology for feature selection classification accuracy evaluation, and its principle will not be repeated.

[0121] After obtaining the fitness of each individual particle at the t-th optimization search, the maximum fitness can be screened out and compared with the global optimal fitness obtained in the previous t-1 times. If the maximum fitness is greater than the historical global optimal fitness, it is used as the global optimal fitness. Otherwise, the historical global optimal fitness is maintained, that is, the historical global optimal fitness is used as the global optimal fitness of this time; then, it is determined whether it is greater than or equal to the first fitness threshold, so as to determine whether to continue the optimization process based on the judgment result; wherein the aforementioned process can be but is not limited to as shown in the following steps S43 and S44.

[0122] S43. Based on the calculated fitnesses, determine the global optimal fitness.

[0123] S44. Determine whether the global optimal fitness is greater than or equal to the first fitness threshold, or whether t reaches the maximum number of optimization searches; in this embodiment, the first fitness threshold and the maximum number of optimization searches can be specifically set according to actual use and are not limited here.

[0124] In this embodiment, if the aforementioned global optimal fitness is less than the first fitness threshold or t does not reach the maximum number of optimization times, at this time, it is necessary to update the position of each individual particle; wherein, this embodiment adopts an improved position update method to perform position update, and its specific process is: by introducing adaptive weights, to increase the global optimization ability and convergence speed of the algorithm; optionally, the position update process of each individual particle can be but is not limited to as shown in the following step S45.

[0125] S45. If not, calculate the position update weight, and based on the position update weight, update the position vector of each individual particle, so as to obtain a number of updated individual particles after the position vector is updated; in specific implementation, the position update weight can be calculated by, for example but not limited to, the following formula (3);

[0126]

[0127] In the above formula (3), ω represents the position update weight, ω max represents the maximum weight, ω min represents the minimum weight, and T represents the maximum number of optimization attempts. In this embodiment, the aforementioned maximum weight and minimum weight are both preset values. It can be seen from the aforementioned formula (3) that in the early stage of optimization, the position update weight is relatively large, and as the number of optimization attempts increases, the position update weight gradually decreases. Thus, in this embodiment, a larger weight is used to expand the algorithm search range when the iteration is just started, and the weight is reduced at the end of the iteration to improve the convergence accuracy. Thus, based on the position update weight, the global optimization capability and convergence speed of the algorithm can be improved, thereby ensuring the accuracy and efficiency of optimization.

[0128] After calculating the position update weight of this optimization search, the position vector of each individual particle can be updated based on the position update weight. In this embodiment, any individual particle is taken as an example for explanation, where the example can be but is not limited to using the following formulas (10) and (11) to update the position vector of any individual particle.

[0129] L u ′=L u +v u ′ (10)

[0130] v u ′=ω×v u +σ1ξ1(q u -L u )+σ2ξ2(g u -L u ) (11)

[0131] In the above formula (10), L u ′ represents the uth element in the updated position vector of any individual particle, L u represents the uth element in the position vector of any individual particle at the tth optimization, v u ′ represents the u-th element in the updated velocity vector of any individual particle.

[0132] In the above formula (12), v u represents the uth element in the velocity vector of any individual particle at the tth optimization, q u represents the uth element in the optimal position vector of any individual particle in the previous t optimization processes (i.e., the historical optimal position of any individual particle, that is, the position vector corresponding to the maximum fitness in the previous t optimization processes), g u represents the u-th element in the position vector corresponding to the global optimal particle (i.e., the individual particle with the largest fitness in the first t iterations), σ1, σ2, ξ1, ξ2 represent the first learning factor, the second learning factor, the third random number, and the fourth random number respectively; wherein the third random number and the third random number are also between [0,1], and u = 1, 2, ..., U, where U represents the total number of elements in the position vector of any individual particle.

[0133] In this way, based on the above formula (10) and formula (11), the position update of each individual particle can be completed; then, the position mapping of the individual particles can be performed, so that the position vectors of all individual particles are mapped into binary vectors for subsequent fitness calculation; wherein, the particle mapping process can be but is not limited to as shown in the following step S46.

[0134] S46. Perform position mapping processing on each updated individual particle to obtain a number of mapped individual particles, and use the several mapped individual particles to form a new particle population; in this embodiment, the position mapping process of the updated individual particles can refer to the aforementioned steps S32a~S32c, which will not be repeated here.

[0135] After completing the position mapping of each updated individual particle, it is necessary to recalculate the fitness of each updated individual particle so as to re-determine the global optimal fitness based on the newly calculated fitness, that is, if the maximum fitness in this optimization process is greater than the historical global optimal fitness, the maximum fitness of this time is used as the global optimal fitness, otherwise, the historical global optimal fitness is not updated; then, the global optimal fitness is compared with the first fitness threshold, and the position is updated according to the comparison result; in this way, the aforementioned process is continuously repeated until the optimization iteration stop condition is met, and the optimal individual particle can be obtained; wherein, the iterative optimization process can be but is not limited to as shown in the following step S47.

[0136] S47. Update the particle population to the new particle population and increment t by 1, and recalculate the fitness of each individual particle in the particle population at the t-th optimization search until the maximum fitness is greater than or equal to the first fitness threshold, or when t reaches the maximum number of optimization searches, the individual particle corresponding to the maximum fitness is used as the optimal individual particle.

[0137] In this way, through the aforementioned steps S41 to S47, the optimization process of the particle population can be completed, thereby obtaining the optimal individual particle; then, the second blasting fragmentation influence index can be determined based on the position vector corresponding to the optimal individual particle; wherein, as mentioned above, the position vector is a binary vector, and an element of 1 indicates that the initial second blasting fragmentation influence index corresponding to the element is selected, and a value of 0 indicates that it is not selected; therefore, the initial second blasting fragmentation influence index corresponding to the element with a value of 1 in the position vector corresponding to the optimal individual particle can be used as the second blasting fragmentation influence index.

[0138] After obtaining the first blasting fragmentation influence index and the second blasting fragmentation influence index, the two can be combined to obtain the optimal blasting fragmentation influence index, and the process can be but is not limited to as shown in the following step S5.

[0139] S5. Determine the optimal blasting fragmentation influence index based on the first blasting fragmentation influence index and the second blasting fragmentation influence index; in this embodiment, the union of the first blasting fragmentation influence index and the second blasting fragmentation influence index is used as the optimal blasting fragmentation influence index; of course, when there are repeated indicators in the union of the two, one of the repeated indicators is retained.

[0140] In this way, after obtaining the optimal blasting fragmentation influence index through the aforementioned step S5, training data can be obtained based on the optimal blasting fragmentation influence index, and model training can be performed to obtain a blasting fragmentation prediction model; then, the blasting fragmentation prediction result of the specified mine can be obtained based on the blasting fragmentation prediction model; wherein, the model training process can be but is not limited to as shown in the following steps S6 and S7.

[0141] S6. Based on the optimal blasting block size influencing index, historical blasting data is obtained; in the present embodiment, it is assumed that the optimal blasting block size influencing index is: borehole diameter, blasting hole spacing, minimum resistance line, blasthole blockage length, explosive consumption, elastic modulus and rock size; then, the data corresponding to the aforementioned 7 optimal blasting block size influencing indexes and the blasting block size of each blasted mine (which serves as label data) are obtained from the historical data of each blasted mine, so that the aforementioned acquired data is used as historical blasting data; of course, if the optimal blasting block size influencing index is other indexes, then the corresponding historical blasting data acquisition process is the same as the aforementioned example, which will not be repeated here.

[0142] After obtaining the historical blasting data corresponding to the optimal blasting block size influence index, the model can be trained based on the historical blasting data. The training process can be, but is not limited to, as shown in the following step S7.

[0143] S7. Use a neural network optimization algorithm, and use the historical blasting fragmentation influence data in the historical blasting data as input and the historical mine blasting fragmentation prediction value as output to train the neural network model, so as to obtain a blasting fragmentation prediction model after the training is completed; in the specific implementation, the neural network model can be, but not limited to, an SVM (support vector machine) model or a BP neural network model; at the same time, this embodiment uses an improved gray wolf algorithm to train the model, and the process can be, but not limited to, the following steps S70 to S79.

[0144] S70. Obtain the model parameters of the neural network model, and construct a gray wolf population based on the model parameters, wherein the gray wolf population includes a number of gray wolves, and the position of any gray wolf is used to characterize a set of model parameters; in the present embodiment, taking the SVM model as an example, the corresponding model parameters are the penalty coefficient and the kernel parameter; therefore, multiple sets of model parameters can be set (each set of model parameters includes a penalty coefficient and a kernel parameter), and then, a number of gray wolves with the same number of model parameter groups are set (i.e., one gray wolf corresponds to a set of model parameters); then, the position of each gray wolf is used to characterize the corresponding model parameters; in this way, based on the aforementioned method, a gray wolf population can be constructed; based on this, the hunting process of the gray wolves in the gray wolf population can be used to optimize the optimal model parameters of the SVM model; specifically, the gray wolf optimization process can be but is not limited to the following steps S71 to S77.

[0145] S71. Initialize the number of iterations s to 1, and calculate the fitness of each gray wolf in the gray wolf population at the s-th iteration, wherein the fitness of any gray wolf is calculated based on the label value and output data corresponding to the input data of the target neural network corresponding to the any gray wolf, the input data of the target neural network is the historical blasting data, and the model parameters of the target neural network are the model parameters corresponding to the position of any gray wolf; in this embodiment, it has been explained above that each gray wolf represents a set of model parameters, therefore, the SVM model corresponding to each gray wolf (i.e., the aforementioned target neural network) can be established based on the model parameters corresponding to each gray wolf. ), then, the aforementioned historical blasting data is divided into a training set and a test set, and the training set is used to train the SVM model corresponding to each gray wolf; after the training is completed, the test set is used to test the SVM model to obtain the model prediction value (i.e., output data) and the actual value (i.e., label value); finally, the model prediction value and the actual value are used to obtain the fitness of each gray wolf at the sth iteration; optionally, the mean square error (MSE) can be calculated using the model prediction value and the actual value, and then 1 / MSE is used as the fitness; thus, the greater the fitness of any gray wolf, the smaller the mean square error of its corresponding SVM model, and the more accurate its model prediction result.

[0146] After calculating the fitness of each gray wolf at the sth iteration, three leader wolves in the gray wolf population can be selected based on the fitness of the gray wolves, so as to update the positions of the remaining gray wolves based on the three leader wolves; wherein, the process of determining the three leader wolves can be but is not limited to as shown in the following step S72.

[0147] S72. Sort each gray wolf in descending order of fitness, and use the top three gray wolves as α wolf, β wolf and δ wolf in turn, and use the gray wolves ranked after the third as follower wolves; in this embodiment, it is equivalent to using the top three gray wolves with the largest fitness as leader wolves, and then use the remaining gray wolves as bottom wolves (that is, follower wolves); then, it can be determined whether the fitness of the aforementioned α wolf is greater than the second fitness threshold, or whether s reaches the maximum number of iterations; so that the positions of each gray wolf can be updated based on the judgment result, wherein the aforementioned judgment process can be but is not limited to as shown in the following step S73.

[0148] S73. Determine whether the fitness of wolf α is greater than or equal to the second fitness threshold, or whether s has reached the maximum number of iterations; in this embodiment, the second fitness threshold and the maximum number of iterations can be specifically set according to actual use, and are not specifically limited here; wherein, if the fitness of wolf α is less than the second fitness threshold, or s has not reached the maximum number of iterations, then it is necessary to update the positions of each gray wolf, wherein this embodiment introduces a nonlinear update factor to solve the problem of the traditional gray wolf algorithm being prone to falling into local optimality; optionally, the calculation process of the gray wolf update factor can be but is not limited to as shown in the following step S74.

[0149] If not, calculate the gray wolf update factor and the distance between each follower wolf and the α wolf, β wolf and δ wolf. In specific implementation, the gray wolf update factor can be calculated by, for example but not limited to, the following formula (4).

[0150]

[0151] In the above formula (4), θ represents the grey wolf update factor, and S represents the maximum number of iterations.

[0152] After obtaining the gray wolf update factor based on the above formula (4), it is necessary to calculate the distance between each follower wolf and the α wolf, β wolf and δ wolf; wherein, taking any follower wolf as an example, the above process is explained, wherein, for example, the following formulas (12), (13) and (14) can be used, but are not limited to, to calculate the distance between any of the above follower wolves and the α wolf, β wolf and δ wolf.

[0153]

[0154] In the above formula (12), D α represents the distance between any following wolf and wolf α, C α represents the second coefficient, represents the position of wolf α (it is a vector, each element in the vector corresponds to a model parameter), represents the position of any wolf following, | | represents the modulo operation;

[0155] In the above formula (13), D β represents the distance between any following wolf and β wolf, C β represents the third coefficient, Indicates the position of β wolf.

[0156] Similarly, in the above formula (14), D δ represents the distance between any following wolf and δ wolf, C δ represents the fourth coefficient, Denotes the position of δ wolf.

[0157] In addition, in this embodiment, the calculation formula of the second coefficient is: 2×ι, where ι represents a fifth random number between [0,1]; of course, the calculation method of the third coefficient and the fourth coefficient is the same, which will not be repeated here.

[0158] Therefore, after calculating the gray wolf update factor and the distance between each follower wolf and α wolf, β wolf and δ wolf based on the above formula, the position of each follower wolf and the three leading wolves can be updated, and the process can be but not limited to as shown in the following step S75.

[0159] S75. Update the positions of wolf α, wolf β and wolf δ, and update the positions of each follower wolf according to the distance between each follower wolf and wolf α, wolf β and wolf δ, and the gray wolf update factor; in the specific implementation, any follower wolf is taken as an example to illustrate its position update process, which can be but not limited to the following steps S75a to S75d.

[0160] S75a. For any follower wolf, based on the gray wolf update factor, the first coefficient is calculated using the following formula (5).

[0161] A=2×θ×b1-θ (5)

[0162] In the above formula (5), A represents the first coefficient, and b1 represents a second random number between [0, 1].

[0163] In this embodiment, the number of second random numbers is three, that is, α wolf, β wolf and δ wolf each correspond to a second random number; thus, the three second random numbers are substituted into the aforementioned formula (5) respectively, and three first coefficients can be obtained, that is, α wolf, β wolf and δ wolf each correspond to a first coefficient; and after obtaining the first coefficients, the distance between any of the aforementioned following wolves and α wolf, β wolf and δ wolf can be combined to calculate the trend amount of any of the following wolves moving toward α wolf, β wolf and δ wolf; wherein, the calculation process of the trend amount can be but is not limited to as shown in the following step S75b.

[0164] S75b. Based on the first coefficient and the distance between any follower wolf and α wolf, β wolf and δ wolf, respectively calculate the trend amount of any follower wolf moving toward α wolf, β wolf and δ wolf; in this embodiment, the trend amount of any follower wolf moving toward α wolf is taken as an example for explanation; wherein, the example can be but not limited to using the following formula (15) to calculate the trend amount of any follower wolf moving toward α wolf.

[0165]

[0166] In the above formula (15), X α,jrepresents the trend amount of any following wolf moving toward α wolf, and A represents the first coefficient corresponding to α wolf; of course, the calculation principle of the trend amount of any following wolf moving toward β wolf and δ wolf is the same, which will not be elaborated here; therefore, after calculating the trend amount of any following wolf moving toward α wolf, β wolf and δ wolf based on the aforementioned formula (15), the corresponding moving trend weight can be calculated based on the aforementioned trend amounts; wherein, the calculation process of the moving trend weight can be but is not limited to as shown in the following step S75c.

[0167] S75c. Calculate the movement trend weight between any of the following wolves and the α wolf, β wolf and δ wolf according to the trend amount of any of the following wolves moving toward the α wolf, β wolf and δ wolf.

[0168] In a specific implementation, for example, but not limited to, the following formulas (6), (7) and (8) may be used in sequence to calculate the movement trend weight between any follower wolf and the α wolf, the β wolf and the δ wolf.

[0169]

[0170] In the above formula (6), W α,j represents the moving trend weight between any following wolf and α wolf, X α,j ,X β,j ,X δ,j represents the trend amount of any following wolf moving toward wolf α, the trend amount of moving toward wolf β, and the trend amount of moving toward wolf δ, respectively. represents the learning factor; among them, for example

[0171] In the above formula (7), W β,j Represents the moving trend weight between any follower wolf and β wolf.

[0172] In the above formula (8), W δ,j Represents the moving trend weight between any following wolf and delta wolf.

[0173] In this way, after calculating the corresponding movement trend weight based on the aforementioned formulas (6)-(8), the position of any following wolf can be updated according to the movement trend weight between any following wolf and α wolf, β wolf and δ wolf. The process can be but is not limited to as shown in the following step S75d.

[0174] S75d. Update the position of any follower wolf using the movement trend weights between any follower wolf and α wolf, β wolf and δ wolf, so as to obtain an updated follower wolf after the position is updated. In this embodiment, the position of any follower wolf can be updated by, but is not limited to, the following formula (9).

[0175]

[0176] In the above formula (9), represents the updated position of any following wolf, represents the vector representation of the trend amount of any following wolf moving towards wolf α; it can be seen from the above formula (15) that the trend amount is actually the modulus of the vector. Therefore, when updating the position, the trend amount needs to be converted into a vector form, that is,

[0177] Therefore, the position update of each follower wolf can be completed through the above formula; then, the position update of the three leading wolves can be performed, among which, for wolf δ, it is led by wolf α and wolf β, and its update method is as follows:

[0178]

[0179] In the above formula (16), represents the updated position of δ wolf, represents the position of δ wolf, P δ represents the intermediate parameter, where P δ It is calculated using the following formula (17):

[0180]

[0181] In the above formula (17), φ represents a sixth random number between [0, 1].

[0182] For wolf β, it accepts the leadership of wolf α, and its position update process is:

[0183]

[0184] In formula (18), represents the updated position of β wolf, γ represents a constant used to define the shape of the logarithmic spiral, represents the seventh random number between [-1,1]; thus, the above formula can make the β wolf approach the α wolf in a spiral motion.

[0185] Similarly, wolf α is not guided by other wolves, so it can, but is not limited to, use the following formula (19) to update its position.

[0186]

[0187] In the above formula (19), Γ represents the step size control factor, Levy represents the random step size that obeys the Levy distribution, Y() represents the fitness function, rand represents the eighth random number between [0,1] and is generated when the α wolf position is updated, and Ψ represents the update probability.

[0188] In this way, based on the aforementioned steps S75a to S75d, the position update of α wolf, β wolf, δ wolf and each following wolf can be completed; then, each updated gray wolves can be used to form a new gray wolf population, and the process can be but not limited to as shown in the following step S76.

[0189] S76. Use the updated positions of α wolf, β wolf, δ wolf and each follower wolves to form a new gray wolf population; in this embodiment, after obtaining the new gray wolf population, it is necessary to recalculate the fitness of each gray wolf in the new gray wolf population, and re-execute the aforementioned steps S72 to S75 until the fitness of the found α wolf is greater than the second fitness threshold, or the number of iterations reaches the maximum number of iterations; wherein, the optimization iterative process can be but is not limited to as shown in the following step S77.

[0190] S77. Add 1 to s and update the gray wolf population to the new gray wolf population, and recalculate the fitness of each gray wolf in the gray wolf population at the s-th iteration, until the fitness of wolf α is greater than or equal to the second fitness threshold, or s reaches the maximum number of iterations, and the model parameters represented by the position of wolf α are used as the optimal model parameters.

[0191] Based on the aforementioned steps S72 to S77, the improved grey wolf algorithm can be used to determine the optimal model parameters of the neural network model, and then, based on the model parameters, the neural network model can be optimized to obtain the optimized neural network model; finally, the blasting fragmentation prediction model can be obtained based on the optimized neural network model; wherein, the model optimization process can be but is not limited to as shown in the following step S78.

[0192] S78. Based on the optimal model parameters, the neural network model is optimized to obtain an optimized neural network model, and the blasting fragmentation prediction model is obtained using the optimized neural network model. In the present embodiment, it is equivalent to directly adjusting the model parameters of the neural network model to the aforementioned optimal model parameters. Then, the historical blasting data is used to train the model, so that after the training is completed, the blasting fragmentation prediction model is obtained. At the same time, each time the fitness is calculated iteratively, the target neural network of each gray wolf is trained. Therefore, the target neural network corresponding to the α wolf that meets the iteration stop condition can also be directly used as the aforementioned blasting fragmentation prediction model.

[0193] After completing the model training and obtaining the blasting fragmentation prediction model, the blasting fragmentation of a specified mine can be predicted based on it. The process can be but is not limited to the following step S8.

[0194] S8. Utilize the blasting fragmentation prediction model to obtain the blasting fragmentation prediction result of the specified mine; in the present embodiment, it is equivalent to obtaining the data corresponding to the optimal blasting fragmentation influencing index of the specified mine (i.e. the drill hole diameter, blasting hole spacing, minimum resistance line, blast hole blockage length, explosive unit consumption, elastic modulus and rock size of the specified mine); then, input the aforementioned data into the blasting fragmentation prediction model to obtain the blasting fragmentation prediction result.

[0195] Thus, through the blasting fragmentation prediction method based on the neural network optimization algorithm described in detail in the aforementioned steps S1 to S8, the present invention comprehensively determines the index data most relevant to the blasting fragmentation based on the features selected by the random forest and particle swarm optimization algorithms, and then, based on the determined most relevant index data, performs model training to obtain a blasting fragmentation prediction model; finally, based on the blasting fragmentation prediction model, a blasting fragmentation prediction result of a specified mine can be obtained; in this way, the present invention realizes feature selection based on data-driven, thereby avoiding the problem that traditional technologies are easily affected by human subjective experience, based on this, improves the accuracy and objectivity of feature selection, thereby increasing the accuracy of subsequent blasting fragmentation predictions, and therefore, is very suitable for large-scale application and promotion.

[0196] like Figure 2 As shown, the second aspect of this embodiment provides a hardware device for implementing the blasting fragmentation prediction method based on the neural network optimization algorithm described in the first aspect of the embodiment, including:

[0197] The acquisition unit is used to obtain the blasting block size impact data corresponding to the specified mine.

[0198] The feature selection unit is used to use a random forest algorithm to perform feature selection processing on the blasting block size influence data to screen out a first blasting block size influence index from the blasting block size influence data.

[0199] A feature selection unit is used to construct a particle population based on the blasting block size influence data, wherein the particle population includes a number of individual particles, and each element in the position vector of any individual particle corresponds to an indicator data in the blasting block size influence data. The position vector of any individual particle is a binary vector, and any element in the position vector corresponding to any individual particle is 1, indicating that the indicator data corresponding to the any element is selected as an initial second blasting block size influence indicator, and any element in the position vector corresponding to any individual particle is 0, indicating that the indicator data corresponding to the any element is not selected.

[0200] The feature selection unit is used to use the improved particle swarm optimization algorithm to perform optimization processing on the particle population to obtain the optimal individual particles, and use the optimal individual particles to determine the second blasting block size influence index from the blasting block size influence data.

[0201] The feature selection unit is also used to determine the optimal blasting fragmentation influence index based on the first blasting fragmentation influence index and the second blasting fragmentation influence index.

[0202] The training unit is used to obtain historical blasting data based on the optimal blasting block size impact index.

[0203] The training unit is also used to adopt a neural network optimization algorithm, and use the historical blasting fragmentation impact data in the historical blasting data as input and the blasting fragmentation prediction value of the historical mine as output to train the neural network model, so as to obtain a blasting fragmentation prediction model after the training is completed.

[0204] The fragmentation prediction unit is used to obtain the blasting fragmentation prediction result of the specified mine by using the blasting fragmentation prediction model.

[0205] The working process, working details and technical effects of the device provided in this embodiment can be found in the first aspect of the embodiment and will not be described in detail here.

[0206] like Figure 3 As shown, the third aspect of this embodiment provides another device for predicting blasting fragmentation based on a neural network optimization algorithm. Taking the device as an electronic device as an example, it includes: a memory, a processor and a transceiver that are communicatively connected in sequence, wherein the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute the method for predicting blasting fragmentation based on a neural network optimization algorithm as described in the first aspect of the embodiment.

[0207] For example, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in first-out memory (FIFO) and / or first-in last-out memory (FILO), etc. Specifically, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). At the same time, the processor may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state.

[0208] In some embodiments, the processor may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. For example, the processor may not be limited to a microprocessor of the STM32F105 series, a reduced instruction set computer (RISC) microprocessor, an X86 architecture processor, or a processor with an integrated embedded neural network processing unit (NPU); the transceiver may be, but is not limited to, a wireless fidelity (WIFI) wireless transceiver, a Bluetooth wireless transceiver, a general packet radio service technology (General Packet Radio Service, GPRS) wireless transceiver, a ZigBee protocol (a low-power local area network protocol based on the IEEE802.15.4 standard, ZigBee) wireless transceiver, a 3G transceiver, a 4G transceiver and / or a 5G transceiver, etc. In addition, the device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0209] The working process, working details and technical effects of the electronic device provided in this embodiment can be found in the first aspect of the embodiment and will not be described in detail here.

[0210] The fourth aspect of this embodiment provides a storage medium storing instructions for the blasting fragmentation prediction method based on a neural network optimization algorithm as described in the first aspect of the embodiment, that is, the storage medium stores instructions, and when the instructions are executed on a computer, the blasting fragmentation prediction method based on a neural network optimization algorithm as described in the first aspect of the embodiment is executed.

[0211] The storage medium refers to a carrier for storing data, which may include but is not limited to a floppy disk, a CD, a hard disk, a flash memory, a USB flash drive and / or a memory stick, and the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0212] The working process, working details and technical effects of the storage medium provided in this embodiment can be found in the first aspect of the embodiment and will not be described in detail here.

[0213] The fifth aspect of this embodiment provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the blasting fragmentation prediction method based on a neural network optimization algorithm as described in the first aspect of the embodiment, wherein the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0214] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting blasting fragmentation based on a neural network optimization algorithm, characterized in that: include: Obtain the blasting block size impact data corresponding to the specified mine; Using a random forest algorithm, feature selection processing is performed on the blasting block size influence data to screen out a first blasting block size influence index from the blasting block size influence data; Based on the blasting block size influence data, a particle population is constructed, wherein the particle population includes a number of individual particles, each element in the position vector of any individual particle corresponds to an index data in the blasting block size influence data, the position vector of any individual particle is a binary vector, and any element in the position vector corresponding to any individual particle is 1, indicating that the index data corresponding to the any element is selected as an initial second blasting block size influence index, and any element in the position vector corresponding to any individual particle is 0, indicating that the index data corresponding to the any element is not selected; Using an improved particle swarm optimization algorithm, the particle population is optimized to obtain optimal individual particles, and using the optimal individual particles to determine a second blasting block size influence index from the blasting block size influence data; Determine the optimal blasting fragmentation influence index according to the first blasting fragmentation influence index and the second blasting fragmentation influence index; Based on the optimal blasting block size influence index, obtaining historical blasting data; A neural network optimization algorithm is used, and the historical blasting block size influence data in the historical blasting data is used as input, and the blasting block size prediction value of the historical mine is used as output to train the neural network model, so as to obtain a blasting block size prediction model after the training is completed; The blasting fragmentation prediction model is used to obtain the blasting fragmentation prediction result of the designated mine.

2. The method according to claim 1, characterized in that Based on the blasting block size impact data, a particle population is constructed, including: According to the blasting block size influence data, a plurality of initial particles are constructed, wherein each element in the position vector of any initial particle corresponds to an indicator data in the blasting block size influence data; Performing initialization processing on each initial particle, and after the initialization processing, performing position mapping processing on each initial particle, so as to obtain individual particles corresponding to each initial particle after the position mapping processing; Using individual particles corresponding to each initial particle to form the particle population; Among them, using the improved particle swarm optimization algorithm to optimize the particle population to obtain the optimal individual particle includes: Obtaining a fitness function, wherein the fitness function is used to characterize the selection accuracy of the initial second blasting block size influence index selected based on any individual particle, and the greater the fitness of any individual particle, the higher the selection accuracy of the initial second blasting block size influence index selected based on the any individual particle; Initialize the optimization number t to 1, and calculate the fitness of each individual particle in the particle population at the tth optimization; Based on the calculated fitnesses, the global optimal fitness is determined; Determine whether the global optimal fitness is greater than or equal to the first fitness threshold, or whether t reaches the maximum number of optimization times; If not, a position update weight is calculated, and based on the position update weight, the position vector of each individual particle is updated to obtain a number of updated individual particles after the position vector is updated; Performing position mapping processing on each updated individual particle to obtain a number of mapped individual particles, and using the number of mapped individual particles to form a new particle population; The particle population is updated to the new particle population, t is incremented by 1, and the fitness of each individual particle in the particle population at the t-th optimization search is recalculated until the global optimal fitness is greater than or equal to the first fitness threshold, or t reaches the maximum number of optimization searches, and the individual particle corresponding to the global optimal fitness is used as the optimal individual particle; Correspondingly, using the optimal individual particles, determining a second blasting fragmentation influence index from the blasting fragmentation influence data includes: The initial second blasting fragmentation influence index corresponding to the element with a value of 1 in the position vector corresponding to the optimal individual particle is used as the second blasting fragmentation influence index.

3. The method according to claim 2, characterized in that Performing position mapping processing on each initial particle to obtain individual particles corresponding to each initial particle after the position mapping processing, including: For any initial particle, the mapping value of each element in the position vector corresponding to the initial particle is calculated according to the following formula (1); In the above formula (1), f(r) represents the mapping value of the rth element in the position vector corresponding to any initial particle, x r represents the rth element in the position vector corresponding to any of the initial particles, wherein r=1, 2, ..., R, and R represents the total number of elements in the position vector corresponding to any of the initial particles; Generate a first random number for any of the initial particles, wherein the value interval of the first random number is [0, 1]; Based on the first random number and the mapping value of each element in the position vector corresponding to any initial particle, the position vector of any initial particle is mapped to obtain the individual particle corresponding to any initial particle after the mapping process, wherein, when performing the mapping process, for any element in the position vector corresponding to any initial particle, if the first random number is greater than or equal to the mapping value of any element, the any element is set to 1; if the first random number is less than the mapping value of any element, the any element is set to 0.

4. The method according to claim 2, characterized in that: The fitness function is: In formula (2), F represents the fitness function, κ represents the selection accuracy, Z represents the number of initial second blasting block size influence indicators determined based on individual particles, N represents the total number of indicators in the blasting block size influence data, η1, η2 both represent selection weights, wherein the selection accuracy is obtained by using a specified classification algorithm to classify the initial second blasting block size influence indicators determined based on individual particles; The calculation of the position update weight includes: The position update weight is calculated using the following formula (3); In the above formula (3), ω represents the position update weight, ω max represents the maximum weight, ω min represents the minimum weight, and T represents the maximum number of optimization times.

5. The method according to claim 1, characterized in that A neural network optimization algorithm is adopted, and the historical blasting block size impact data in the historical blasting data is used as input, and the blasting block size prediction value of the historical mine is used as output to train the neural network model, including: Obtaining model parameters of the neural network model, and constructing a gray wolf population based on the model parameters, wherein the gray wolf population includes a plurality of gray wolves, and the position of any gray wolf is used to represent a set of model parameters; Initialize the number of iterations s to 1, and calculate the fitness of each gray wolf in the gray wolf population at the sth iteration, wherein the fitness of any gray wolf is calculated based on the label value and output data corresponding to the input data of the target neural network corresponding to the any gray wolf, the input data of the target neural network is the historical blasting data, and the model parameters of the target neural network are the model parameters corresponding to the position of any gray wolf; According to the order of fitness from large to small, each gray wolf is sorted, and the first three gray wolves are regarded as α wolf, β wolf and δ wolf in turn, and the gray wolves ranked after the third place are regarded as follower wolves; Determine whether the fitness of wolf α is greater than or equal to the second fitness threshold, or whether s reaches the maximum number of iterations; If not, calculate the gray wolf update factor and the distance between each follower wolf and α wolf, β wolf and δ wolf; Update the positions of the α wolf, β wolf and δ wolf, and update the positions of the following wolves according to the distances between the following wolves and the α wolf, β wolf and δ wolf, and the gray wolf update factor; Use the updated positions of α wolf, β wolf, δ wolf and each follower wolf to form a new gray wolf population; Increment s by 1 and update the gray wolf population to the new gray wolf population, and recalculate the fitness of each gray wolf in the gray wolf population at the sth iteration, until the fitness of wolf α is greater than or equal to the second fitness threshold, or s reaches the maximum number of iterations, and the model parameters represented by the position of wolf α are used as the optimal model parameters; Based on the optimal model parameters, the neural network model is optimized to obtain an optimized neural network model, and the blasting fragmentation prediction model is obtained using the optimized neural network model.

6. The method according to claim 5, characterized in that Calculate the gray wolf update factor, including: The gray wolf renewal factor is calculated using the following formula (4); In the above formula (4), θ represents the gray wolf update factor, and S represents the maximum number of iterations; Accordingly, updating the position of each follower wolf according to the distance between each follower wolf and the α wolf, the β wolf and the δ wolf, and the gray wolf update factor, includes: For any follower wolf, based on the gray wolf update factor, the first coefficient is calculated using the following formula (5); A=2×θ×b1-θ (5) In the above formula (5), A represents the first coefficient, and b1 represents a second random number between [0, 1]; Based on the first coefficient and the distance between any of the following wolves and the α wolf, the β wolf and the δ wolf, respectively calculating the tendency amount of any of the following wolves to move toward the α wolf, the β wolf and the δ wolf; According to the trend amount of any follower wolf moving toward the α wolf, the β wolf, and the δ wolf, a moving trend weight between any follower wolf and the α wolf, the β wolf, and the δ wolf is calculated; The position of any following wolf is updated by using the movement trend weights between any following wolf and the α wolf, the β wolf and the δ wolf, so as to obtain the updated any following wolf after the position is updated.

7. The method according to claim 6, characterized in that According to the trend amount of any follower wolf moving toward α wolf, β wolf and δ wolf, the moving trend weight between any follower wolf and α wolf, β wolf and δ wolf is calculated, including: The following formulas (6), (7) and (8) are used in sequence to calculate the movement trend weights between any following wolf and the α wolf, the β wolf and the δ wolf; In the above formula (6), W α,j represents the moving trend weight between any following wolf and α wolf, X α,j ,X β,j ,X δ,j represents the trend amount of any following wolf moving toward wolf α, the trend amount of moving toward wolf β, and the trend amount of moving toward wolf δ, respectively. represents the learning factor; In the above formula (7), W β,j represents the moving trend weight between any following wolf and β wolf; In the above formula (8), W δ,j represents the moving trend weight between any following wolf and δ wolf; Accordingly, using the movement trend weights between any of the following wolves and the α wolf, the β wolf, and the δ wolf to update the position of any of the following wolves includes: The following formula (9) is used to update the position of any following wolf; In the above formula (9), represents the updated position of any following wolf, A vector representation of the tendency of any of the following wolves to move toward the alpha wolf.

8. A blasting fragmentation prediction device based on a neural network optimization algorithm, characterized in that: include: An acquisition unit, used for acquiring blasting block size impact data corresponding to a specified mine; A feature selection unit, configured to perform feature selection processing on the blasting block size influence data using a random forest algorithm, so as to screen out a first blasting block size influence index from the blasting block size influence data; A feature selection unit is used to construct a particle population based on the blasting block size influence data, wherein the particle population includes a plurality of individual particles, each element in the position vector of any individual particle corresponds to an index data in the blasting block size influence data, the position vector of any individual particle is a binary vector, and any element in the position vector corresponding to any individual particle is 1, indicating that the index data corresponding to the any element is selected as an initial second blasting block size influence index, and any element in the position vector corresponding to any individual particle is 0, indicating that the index data corresponding to the any element is not selected; A feature selection unit is used to use an improved particle swarm optimization algorithm to perform optimization processing on the particle population to obtain optimal individual particles, and use the optimal individual particles to determine a second blasting block size influence index from the blasting block size influence data; The feature selection unit is further used to determine the optimal blasting fragmentation influence index according to the first blasting fragmentation influence index and the second blasting fragmentation influence index; A training unit, used for acquiring historical blasting data based on the optimal blasting block size influence index; The training unit is further used to adopt a neural network optimization algorithm, and use the historical blasting block size impact data in the historical blasting data as input and the blasting block size prediction value of the historical mine as output to train the neural network model, so as to obtain a blasting block size prediction model after the training is completed; The fragmentation prediction unit is used to obtain the blasting fragmentation prediction result of the specified mine by using the blasting fragmentation prediction model.

9. An electronic device, characterized in that: include: A memory, a processor and a transceiver which are sequentially connected in communication, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute the blasting block size prediction method based on a neural network optimization algorithm as described in any one of claims 1 to 7.

10. A computer program product comprising instructions, characterized in that When the instructions are executed on a computer, the computer is enabled to execute the blasting fragmentation prediction method based on a neural network optimization algorithm as described in any one of claims 1 to 7.

Images