Slope blasting lumpiness prediction method based on Buble Sort-GWO-ELM model and application

Abstract

The invention belongs to the field of slope engineering construction, and provides a slope blasting lumpiness prediction method based on a Buble Sort-GWO-ELM model to solve the problem that the construction quality and cost control are affected due to the fact that the slope blasting lumpiness cannot be accurately predicted in the slope construction process, and the slope blasting lumpiness prediction method comprises the steps that blasting control parameters are monitored; performing data preprocessing; constructing a blasting control parameter data set; the blasting control parameters are screened by using bubble sorting in combination with a square difference algorithm, and characteristic parameters which have the greatest influence on a prediction result are selected; training an extreme learning machine (ELM) model, and performing optimization and verification by using an improved grey wolf optimization algorithm model parameter; determining optimal parameters of an extreme learning machine (ELM) model, and establishing a blasting effect prediction model; and predicting by using the blasting effect prediction model. According to the method, the blasting parameters can be scientifically predicted and reasonably controlled, the accuracy of blasting construction can be improved, and the economical efficiency and safety of the whole project can be improved.

Description

Technical Field

[0001] The invention relates to the technical field of slope engineering construction, and in particular to a slope blasting block size prediction method based on a Bubble Sort-GWO-ELM model and an application thereof. Background Art

[0002] Drilling and Blasting Method is a rock excavation technology commonly used in mining, tunneling and large-scale civil engineering. This method achieves the purpose of excavation by drilling, loading explosives, filling inert materials and detonating to break the rock. Its main steps include drilling, charging, filling, blasting and cleaning evaluation. This method is suitable for various hard rocks, has high flexibility and low cost, so it is widely used in mining, tunneling and foundation engineering. With the development of smooth blasting technology and blasting vibration control technology, more and more rock cutting slopes are also using drilling and blasting as the main construction method.

[0003] During the construction of rocky road cutting slopes, the quality of the blasting design has a crucial impact on the construction quality and cost control. This impact is not only reflected in the blasting effect itself, but also affects the implementation effect of the measures taken to reduce vibration. However, there is currently a lack of methods for predicting the size of slope blasting fragments, which affects the design of blasting design schemes. In order to optimize construction quality and cost control, it is urgent to establish an efficient slope blasting fragmentation prediction model. Summary of the invention

[0004] The purpose of the present invention is to provide a slope blasting block size prediction method based on the Bubble Sort-GWO-ELM model, through which reasonable blasting parameters can be accurately obtained, and the negative impact of blasting on existing structures can be actively reduced. It can also effectively control excessive or insufficient excavation volume to avoid cumulative damage to surrounding rocks. At the same time, it can significantly reduce the consumption of slope support materials and reduce the overall cost of drilling and blasting construction.

[0005] To achieve the above object, the present invention provides a slope blasting fragmentation prediction method based on the Bubble Sort-GWO-ELM model, comprising the following steps:

[0006] Step 1: Monitor blasting control parameter data at the construction site;

[0007] Step 2: Preprocessing the monitored blasting control parameter data;

[0008] Step 3: Use bubble sort combined with square difference algorithm to filter the preprocessed blasting control parameter data, build a blasting control parameter data set based on the filtered blasting control parameter data, and divide the blasting control parameter data set into a training set and a test set;

[0009] Step 4: Normalize the obtained training set and test set;

[0010] Step 5: Based on the normalized training set, the parameters of the extreme learning machine ELM model are optimized using the improved gray wolf optimization algorithm to determine the optimal parameters of the optimized extreme learning machine ELM model;

[0011] Step 6: Optimize the extreme learning machine ELM model with the optimal parameters, test the optimized extreme learning machine ELM model on the test set, count the output of the prediction of the test set samples, evaluate the optimal parameters, and establish a blasting effect prediction structure model based on the optimal parameters;

[0012] Step 7: Use the blasting effect prediction model to predict the slope blasting fragmentation and output the prediction results.

[0013] Furthermore, the specific steps of step 3 include:

[0014] Step 3.1: Define the n monitored blasting control parameter data as an array A containing n elements. Each element in array A corresponds to a data to be screened and sorted. Each element X in array A is i It includes six types of blasting control parameters, namely tensile strength, blasting resistance line, step breaking angle, explosive consumption, hole spacing and row spacing;

[0015] Step 3.2: Sort the n elements X in array A by using the bubble sort algorithm. i Sorting is performed, and based on the position of the elements in each round of sorting, different combinations of six types of blasting control parameters are achieved;

[0016] Step 3.3: Calculate the squared difference of each of the different combinations according to the following formula:

[0017]

[0018] Where N is the number of data points in each combination, X i is the actual value monitored, is the mean of the data;

[0019] Step 3.4: Repeat steps 3.2 and 3.3 until the elements in array A are arranged in ascending order, and then obtain the square differences of various combinations in each round of bubble sorting;

[0020] Step 3.5: By comparing the square differences of various combinations, the blasting control parameter data corresponding to the combination with the smallest square difference value is selected;

[0021] Step 3.6: Based on the blasting control parameter data screened in step 3.5, a blasting control parameter data set is established, and the blasting control parameter data set is divided into a training set and a test set.

[0022] Furthermore, the specific steps of step 5 include:

[0023] Step 5.1: Obtain the normalized training set data;

[0024] Step 5.2: Initialize the number of hidden layer neurons L and activation function of the extreme learning machine ELM model;

[0025] Step 5.3: Randomly initialize the input weight matrix W and the hidden layer bias b, and optimize the parameters W and b using the improved gray wolf optimization algorithm based on the data of the training set;

[0026] Step 5.4: Based on the optimized parameters W and b, calculate the hidden layer output matrix H and determine the number of hidden layer neurons L;

[0027] Step 5.5: Output the optimal parameters of the current extreme learning machine ELM model, including the number of hidden layer neurons L, W and b.

[0028] Furthermore, the improved gray wolf optimization algorithm in step 5.3 includes the following steps:

[0029] Step 5.3.1: Initialize the coefficient vectors A and C by:

[0030]

[0031] Where a is the convergence factor, which decreases linearly from 2 to 0 as the number of iterations t decreases. 1 and r 2 is a random vector between [0, 1];

[0032] Step 5.3.2: Calculate the fitness value of each gray wolf in the wolf pack, mark the three wolf leaders with the largest fitness values ​​in the wolf pack as α, β, and δ, respectively, and mark the remaining gray wolves as ω. Calculate the position of the gray wolves using the following formula to surround the prey:

[0033] X(t+1)=Ω·(X p (t)-A·D)

[0034] D=|C·X p (t)-X(t)|

[0035] Among them, t is the current iteration number, X p is the position vector of the current prey, X(t) is the position vector of the t-th generation leader, and Ω is the adaptive weight factor;

[0036] and

[0037] Step 5.3.3: Update the positions of the three alpha wolves in the wolf pack using the following formula:

[0038]

[0039] in, and Respectively represent the distances between α wolf, β wolf and δ wolf and other individuals; and Represent the current positions of α wolf, β wolf and δ wolf respectively; and is a random number, is the current location of the individual gray wolf;

[0040] According to the positions of α, β and δ, the positions of all gray wolf individuals are updated by the following gray wolf individual position update formula:

[0041]

[0042] in, They represent the adjusted position of ω wolf affected by α wolf, β wolf and δ wolf respectively;

[0043]

[0044] in, Indicates the latest updated position of the gray wolf;

[0045] Step 5.3.4: Determine whether the maximum number of iterations has been reached. If so, output the optimal solution. If not, return to step 5.3.2.

[0046] Furthermore, the specific operation steps of step 5.4 include:

[0047] Step 5.4.1: For each input sample x i , the output h of the i-th hidden layer node is obtained according to the following formula i :

[0048] h i =g(w i ,b i ,x)=g(w i x+b i ),w i ∈R D ,b i ∈R

[0049] Among them, w i represents the input weight vector; b i represents the bias of the i-th neuron; x represents the input sample; RD represents D-dimensional real space; g(w i ,b i ,x) represents the activation function;

[0050] Step 5.4.2: Based on the hidden layer output vectors of all samples, form the hidden layer output matrix H:

[0051] H=[h 1 (x),...,h L (x)]

[0052] Where L represents the number of neurons in the hidden layer;

[0053] Step 5.4.3: The output of the single hidden layer feedforward neural network ELM is obtained according to the following formula:

[0054]

[0055] Among them, β is the output weight between the hidden layer and the output layer;

[0056] Step 5.4.4: Establish the optimization objective function of the output weight β based on the output weight β:

[0057] min||Hβ-T|| 2 ,β∈R L×m

[0058] Where H represents the output matrix of the hidden layer, T represents the target matrix, which contains the true output label or value of each training sample; m represents the number of training set samples;

[0059] Step 5.4.5: Solve the above equation by minimizing the approximate square error method to obtain the optimal solution β * ;

[0060] Step 5.4.6: According to the optimal solution β * Calculate the output matrix of the hidden layer.

[0061] A slope blasting fragmentation prediction system is implemented based on the above-mentioned slope blasting fragmentation prediction method based on the Bubble Sort-GWO-ELM model, which includes a data acquisition module, a data preprocessing module, a parameter screening module, a blasting effect prediction model establishment module and a slope blasting fragmentation prediction module;

[0062] Data acquisition module, used to monitor blasting control parameter data at the construction site;

[0063] A data preprocessing module, used for preprocessing the data collected by the data acquisition module;

[0064] A blasting control parameter data set construction module is used to filter the preprocessed blasting control parameter data by using bubble sort combined with square difference algorithm, construct a blasting control parameter data set based on the filtered blasting control parameter data, and divide the blasting control parameter data set into a training set and a test set;

[0065] A normalization processing module is used to normalize the obtained training set and test set;

[0066] A model optimization module is used to optimize the parameters of the extreme learning machine ELM model based on the normalized training set using the improved grey wolf optimization algorithm to determine the optimal parameters of the optimized extreme learning machine ELM model;

[0067] The blasting effect prediction model establishment module is used to optimize the extreme learning machine ELM model with optimal parameters, test the optimized extreme learning machine ELM model on the test set, calculate the output of the prediction of the test set samples, evaluate the optimal parameters, and establish the blasting effect prediction model based on the optimal parameters;

[0068] The slope blasting fragmentation prediction module is used to predict the slope blasting fragmentation using the blasting effect prediction model and output the prediction results.

[0069] Preferably, the model optimization module includes:

[0070] A normalized data acquisition unit, used to acquire the data of the training set after normalization;

[0071] A model initialization unit, used to initialize the number L of hidden layer neurons and activation function of the extreme learning machine ELM model;

[0072] A parameter optimization unit is used to randomly initialize the input weight matrix W and the hidden layer bias b, and optimize the parameters W and b using the improved gray wolf optimization algorithm based on the data of the training set;

[0073] A hidden layer neuron determination unit is used to calculate the hidden layer output matrix H and determine the number of hidden layer neurons L based on the optimized parameters W and b;

[0074] The optimal parameter output unit is used to output the optimal parameters of the current extreme learning machine ELM model, including the number of hidden layer neurons L, W and b.

[0075] An electronic device comprises: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned slope blasting block size prediction method based on the Bubble Sort-GWO-ELM model.

[0076] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the method for predicting the fragmentation of slope blasting based on the Bubble Sort-GWO-ELM model is implemented.

[0077] Compared with the prior art, the present invention adopts the above-mentioned slope blasting fragmentation prediction method based on the Bubble Sort-GWO-ELM model, which has the following beneficial effects:

[0078] First, the present invention obtains a blasting parameter data set through a Bubble Sort method. The blasting parameter data set includes tensile strength, blasting resistance line, step fracture angle, explosive consumption, hole spacing, and row spacing. These six characteristics are used as evaluation indicators, so that the scheme evaluation result can better reflect the comprehensive effect of the blasting design; at the same time, by designing a prediction model with an improved Grey Wolf Optimization (GWO) algorithm, the characteristics of the Grey Wolf Optimization algorithm such as simple structure, few parameters to be adjusted, and easy implementation are fully utilized, and the slope blasting effect prediction can be effectively completed when used, and it has the advantages of fast convergence speed, high output calculation results and high calculation accuracy;

[0079] Second, the prediction model proposed in the present invention takes GWO as the framework and is screened by Bubble Sort parameters. It has the characteristics of strong generalization ability and high operating efficiency. Compared with other models, the evaluation results of the GWO model are more accurate and efficient.

[0080] In summary, the prediction model can not only improve the accuracy of blasting construction, but also improve the economy and safety of the entire project. Through scientific prediction and reasonable control of blasting parameters, the construction process can be ensured to be smoother, unnecessary waste of resources and potential safety hazards can be reduced, thus providing a strong guarantee for the successful completion of the project.

[0081] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 This is a flow chart of the present invention using the Grey Wolf Algorithm to improve the ELM model.

[0083] Figure 2It is a bar chart of the six characteristic evaluation indicators of the prediction model proposed in the present invention;

[0084] Figure 3 Schematic diagram of the ELM (extreme learning machine) network structure of the prediction model proposed in the present invention.

[0085] Figure 4 The flowchart of the method for establishing the invention is shown in FIG. DETAILED DESCRIPTION

[0086] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art make various changes or modifications to the present invention, and these equivalent forms also fall within the scope limited by the claims attached to the application.

[0087] according to Figure 1-Figure 4 As shown, the present invention provides a slope blasting fragmentation prediction method based on the Bubble Sort-GWO-ELM model, which includes the following steps:

[0088] 1. Establish blasting control parameter data set

[0089] By monitoring the construction site, n blasting control parameters are obtained. Bubble Sort combined with square difference algorithm is used to screen these n blasting control parameters. An initial blasting control parameter data set is constructed based on the screened m blasting control parameters, and the blasting control parameter data set is divided into a training set and a test set.

[0090] Specifically, the specific steps of the Bubble Sort combined with the square difference algorithm for parameter screening are:

[0091] (1) For the n monitored blasting control parameter data, define it as an array A containing n elements, A = {X i}, i = 0, 1, ..., n each element corresponds to a data to be sorted, and each element X in A i (X i ={A j ,B j ,C j ,D j ,E j ,F j}) contains six types of blasting control parameters, namely tensile strength, blasting resistance line, step fracture angle, explosive consumption, hole spacing and row spacing;

[0092] For example, currently 5 blasting control parameter data are monitored, which are defined as an array A = {X2, X4, X5, X3, X1}, where each element Xi, i = 1, 2, 3, 4, 5, Xi = {A, B, C, D, E, F}, that is, each element Xi contains 6 types of blasting control parameters, A, B, C, D, E, F represent tensile strength, blasting resistance line, step fracture angle, explosive consumption, hole spacing and row spacing data respectively;

[0093] (2) Sort n elements X in array A by Bubble Sort i Sorting is performed. Since the bubble sort swaps the elements once in each round, different combinations of the six types of blast control parameters can be achieved, and the square difference (Sum of Squares, SS) of each combination is calculated by the following formula:

[0094]

[0095] Where N is the number of data points, X i is the actual value monitored, is the mean of the data.

[0096] For example, the array A = {X2, X4, X5, X3, X1} containing 5 elements is bubble sorted. When the first round of sorting is performed, the elements are swapped once to obtain the array A. 1 {X 2, X4, X 3, X 1, X 5}, and calculate the current array A 1 Then, perform a second round of sorting and swap the elements again to get array A. 2 {X2, X3, X1, X4, X5}, then calculate the current array A 2 The square difference of , and so on, calculate the array A obtained by the third round of sorting 3 The square difference of the fourth round of sorting, array A 4 The square difference of

[0097] (3) By comparing the square differences of various combinations, the blasting control parameter data corresponding to the combination with the smallest square difference is selected.

[0098] For example, in the sorting process of the array A={X2, X4, X5, X3, X1} containing 5 elements, the square differences of the arrays obtained in each round of sorting are compared respectively, and the parameter data corresponding to the combination with the smallest square difference is screened out.

[0099] The time complexity of the above calculation for bubble sorting of n elements is:

[0100]

[0101] The worst-case and average time complexities of bubble sort are both O(n 2 ). In the optimal case (that is, when the array is already sorted), the time complexity is O(n), because there are no swaps after the first traversal and the algorithm will terminate immediately.

[0102] In the above step (2), n blasting control parameters X i The specific process of bubble sorting is as follows:

[0103] S2.1: Get an array A to be sorted with a length of n;

[0104] S2.2: Initialize variable k. For k=0, 1, ..., n-2, set the flag swapped=False.

[0105] S2.3: For i=1 to nk-1, if A[i]>A[i+1], swap A[i] and A[i+1] and set swapped=True;

[0106] S2.4: Determine whether swapped is equal to False. If so, exit the loop, otherwise return to step S2.3;

[0107] S2.5: Finally, output a sorted array A' of length n in ascending order.

[0108] In summary, the present invention calculates the square differences of different combinations simultaneously during each sorting process through the bubble sort method, and screens out the combination with the smallest square difference by comparing the calculated square differences. Subsequently, the blasting control parameters corresponding to the combination with the smallest square difference can be used as input variables of the adaptive model and the improved gray wolf optimization algorithm.

[0109] Through the above method, various blasting control parameters can be systematically evaluated and screened in the data processing stage, and finally the feature combination with the best model effect can be selected. Using the square error as the evaluation criterion, the feature combination with small volatility and high stability can be effectively screened, thereby improving the accuracy and reliability of the subsequent model.

[0110] 2. Normalize the blasting control parameter data set

[0111] The present invention normalizes the obtained training set and test set, thereby unifying the data range, improving the training speed and stability, and avoiding numerical problems.

[0112] The specific steps of normalization are:

[0113] (1) Unify the input features x jThe dimension of the training set is calculated by the following formula: 1 ,x 2 ,…x j ,…,x n ] T Perform normalization:

[0114]

[0115] Among them, α=0.95, β=0.05; x max 、x min are the maximum and minimum values ​​in the input feature vector respectively;

[0116] (2) The predicted sample data in the test set is calculated by the following formula Perform normalization:

[0117]

[0118] in, Output normalized value for ELM network; b last,1 is the threshold of the first neuron in the hidden layer; W last,1 is the weight; V last is the inter-layer weight; R is the activation function R(x)=max(0,x); U 1 is the input of the hidden layer neurons;

[0119] (3) For the normalized learning sample x 1 And the prediction sample x Ⅱ , the following anti-normalization formula can be used to denormalize the data:

[0120]

[0121] Among them, Y Ⅰ,max =t max ; Y Ⅰ,min =t min ; β is the momentum coefficient, α=0.95.

[0122] 3. Improve the ELM adaptive model and obtain the optimal parameters

[0123] The present invention uses the improved gray wolf optimization algorithm to improve the above-mentioned adaptive model, optimize the optimization effect of the training set, and obtain the optimal parameter result;

[0124] The specific steps include:

[0125] (1) Data preparation and parameter setting: In the previous step, the data has been normalized. In addition, the number L of hidden layer neurons and the activation function of the ELM model need to be determined. The activation function of the present invention selects sigmoid. According to the rule of thumb, the number L of hidden layer neurons is determined through multiple experiments and cross-validation. L is usually a value between the number of neurons in the input layer and the number of neurons in the output layer.

[0126] (2) Randomly initialize the input weight matrix W and hidden layer bias b: The uniqueness of ELM lies in the random initialization of its hidden layer parameters. The input weight matrix W and the hidden layer bias b are randomly generated and usually initialized within a small random range. The present invention uses the gray wolf algorithm to optimize W and b.

[0127] (3) Calculate the hidden layer output matrix H:

[0128] For each input sample x i :

[0129] The output of the hidden layer is the input multiplied by the corresponding weight plus the bias, and then passed through a nonlinear function and the sum of all the node results. H is the ELM nonlinear mapping (hidden layer output matrix), h i is the output of the ith hidden layer node. The output function of the hidden layer node is not unique, and different output functions can be used for different hidden layer neurons. Usually, in practical applications, h i It is expressed as:

[0130] h i =g(w i ,b i ,x)=g(w i x+b i ),w i ∈R D ,b i ∈R (6)

[0131] Among them, w i represents the input weight vector, which is used to connect the input layer with the i-th hidden layer neuron; b i is the bias of the i-th neuron; x represents the input sample; R D represents D-dimensional real space; g(w i ,b i ,x) is the activation function;

[0132] g(w i ,b i , x) is a nonlinear piecewise continuous function that satisfies the ELM universal approximation ability theorem. The present invention uses the sigmoid function, so the g function in the formula is:

[0133]

[0134] The hidden layer output vectors of all samples are combined to form the hidden layer output matrix H. The calculation formula of the hidden layer output matrix H is as follows:

[0135] H=[h 1 (x), ..., h L (x)] (8)

[0136] After passing through the hidden layer, it enters the output layer. According to formula (8), the output of the “generalized” single hidden layer feedforward neural network ELM is:

[0137]

[0138] Among them, β is the output weight between the hidden layer and the output layer. Since W and b have completed the initialization input, it is now necessary to solve the weight β of the output layer.

[0139] In order to obtain β with good effect on the training set, it is necessary to ensure that its training error is minimized. Hβ and sample label T are used to minimize the square error to evaluate the training error, so that the solution with the smallest objective function as shown in formula (10) is the optimal solution. That is, the weight β connecting the hidden layer and the output layer is solved by minimizing the approximate square error. The objective function is as follows:

[0140] min||Hβ-T|| 2 ,β∈R L×m (10)

[0141] Among them, H represents the output matrix of the hidden layer, T represents the target matrix, which contains the true output label or value of each training sample; m represents the number of training set samples, and L represents the number of neurons in the hidden layer.

[0142] The optimal solution to formula (10) is:

[0143] β * =H * T (11)

[0144] Among them, H * is the Moore–Penrose generalized inverse matrix of the matrix H.

[0145] When H * When H is a non-singular matrix, the orthogonal projection method is used to calculate H through equation (12): * :

[0146] H * =(H T H) -1 H T(12);

[0147] Get H * Then, substitute it into formula (11) to get the optimal weight β of the output layer.

[0148] (4) Improved Gray Wolf Optimization Algorithm

[0149] The present invention proposes an improved gray wolf optimization algorithm, which specifically includes:

[0150] S1. Construct a social hierarchy model for gray wolves;

[0151] Specifically, the fitness of each individual in the population is calculated, and the three gray wolves with the best fitness in the wolf pack are marked as α, β, and δ in turn, and the remaining gray wolves are marked as ω. The optimization process of GWO is mainly guided by the best three solutions (i.e., α, β, and δ) in each generation of the population;

[0152] S2. The wolf group surrounded and approached the prey;

[0153] First, initialize the coefficient vectors A and C by:

[0154]

[0155] in:

[0156]

[0157] Where t is the current iteration number, t max is the maximum number of iterations; a is the convergence factor, which decreases linearly from 2 to 0 as the number of iterations decreases. 1 and r 2 It obeys the uniform distribution between [0, 1];

[0158] Secondly, in the traditional GWO, the gray wolf uses the following position update formula to surround the prey during hunting:

[0159]

[0160] Among them, X p is the position vector of the current prey, and X(t) is the position vector of the t-th generation leader.

[0161] The present invention improves the position update step. In the position update process, a dynamic adjustment mechanism is introduced. The above position update formula is improved by introducing an adaptive weight factor Ω, so that the algorithm has a stronger global search capability in the early stage and a stronger local search capability in the later stage, thereby improving the convergence speed and accuracy of the algorithm.

[0162] The improved position update formula is:

[0163] X(t+1)=Ω·(X p (t)-A·D)

[0164] D=|C·X p (t)-X(t)|

[0165] in:

[0166]

[0167] Where t is the current iteration number, Ω is the adaptive weight factor;

[0168] S3. Attack the prey and update the position of the gray wolf;

[0169] In nature, although the hunting process is usually led by the alpha wolf, and other levels of wolves cooperate to surround, hunt and attack the prey, in the process of evolutionary calculation, the position of the prey (optimal solution) is unknown, so in GWO we consider the optimal gray wolf to be α, the second optimal gray wolf to be β, the third optimal gray wolf to be δ, and the remaining gray wolves to be ω. According to the characteristic that α (potential optimal solution), β and δ have more knowledge about the location of the prey, a model is established. In the iterative process, α, β and δ are used to guide the movement of ω, thereby achieving global optimization.

[0170] Using the positions of α, β, and δ, update the positions of the three wolves using the following equations:

[0171]

[0172] in, and Respectively represent the distances between α wolf, β wolf and δ wolf and other individuals; and Represent the current positions of α wolf, β wolf and δ wolf respectively; and is a random number, is the current location of the individual gray wolf.

[0173] According to the positions of wolf α, wolf β and wolf δ, the positions of all gray wolves are updated. The update formula for the position of each individual gray wolf is:

[0174]

[0175] in, They represent the position of ω wolf after being affected by α wolf, β wolf and δ wolf respectively. The average value is taken here, that is:

[0176]

[0177] When the value of the convergence factor a decreases from 2 to 0, the corresponding value of A also changes in the interval [-a, a]: the larger the value of a, the farther the gray wolf will move from the prey, hoping to find a more suitable prey, thus prompting the wolf pack to conduct a global search (|A|>1); if the value of a is smaller, the gray wolf will move closer to the prey, prompting the wolf pack to conduct a local search (|A|<1);

[0178] S4. Search for prey and determine whether the maximum number of iterations has been reached.

[0179] 4. Calculate and evaluate the optimal parameters

[0180] After obtaining the optimal parameters W and b, the ELM model uses the optimal parameters to make predictions on the test set. By statistically analyzing the output of the predictions on the test set samples at this time, the optimal parameter results of the model are evaluated. The evaluation indicators include: MAPE (mean absolute percentage error), MAE (mean absolute error), MSE (mean square error)

[0181] The calculation formula for each evaluation index is:

[0182]

[0183] Among them, m represents the number of samples in the test set, y i is the actual value of the blasting effect evaluation of the i-th sample, It is the calculated blasting effect evaluation prediction value of the i-th sample.

[0184] 5. Build a blasting effect prediction model and predict

[0185] According to the verified optimal parameter results, a blasting effect prediction model is established, and the slope blasting block size prediction is performed using the constructed blasting effect prediction model, which includes the following steps:

[0186] Step 5.1: Use the obtained optimal parameters to retrain the ELM model and establish the final blasting effect prediction model;

[0187] Step 5.2: Use the trained ELM model to predict the slope blasting fragmentation;

[0188] Among them, step 5.1 specifically includes:

[0189] Step 5.1.1: Retrain the model: Use all training data and optimal parameters to retrain the ELM model to ensure that the model fully learns the patterns in the data;

[0190] Step 5.1.2: Model establishment: Use the trained model as the established blasting effect prediction model;

[0191] Step 5.2 specifically includes:

[0192] Step 5.2.1: Input the latest blasting control parameters monitored in real time on site into the trained ELM model;

[0193] Step 5.2.2: Normalize the newly input data to make it consistent with the training data;

[0194] Step 5.2.3: Use the trained ELM model to predict and obtain the prediction result of slope blasting fragmentation;

[0195] Step 5.2.4: Use formula (5) to denormalize the prediction results and convert them back to the original data scale to make the prediction results meaningful.

[0196] Example

[0197] In order to verify the effect of the prediction model proposed in the present invention, the slope blasting fragmentation prediction method based on the Bubble Sort-GWO-ELM model proposed in the present invention is compared with the traditional slope blasting fragmentation prediction method (MLR). Multiple linear regression (MLR) is a traditional method for predicting slope blasting fragmentation. The MLR model assumes that there is a linear relationship between the input parameters and the output results, and the prediction is made by fitting the regression coefficients. The two are compared from the following aspects:

[0198] 1. Data preprocessing and feature selection

[0199] (1) Traditional method (MLR)

[0200] Its feature selection relies on expert experience or simple statistical analysis methods, which may lead to the omission of key features or the addition of redundant features. In addition, the data processing process of MLR is relatively simple and cannot fully utilize the potential information of the data.

[0201] (2) Bubble Sort-GWO-ELM method proposed in the present invention

[0202] It uses Bubble Sort algorithm combined with square difference (SS) for feature selection, which can effectively screen out the features that have the greatest impact on the prediction results. It uses six types of blasting control parameters (tensile strength, blasting resistance line, step fracture angle, explosive unit consumption, hole spacing, row spacing) for combined analysis to ensure that data processing is more scientific and comprehensive, and improve the accuracy and reliability of the model.

[0203] In addition, before building the prediction model, the present invention performs complete data preprocessing of data cleaning, normalization and feature selection on the collected raw data, thereby improving the model performance.

[0204] 2. Model construction

[0205] (1) Traditional method (MLR)

[0206] It uses a simple linear relationship and cannot capture the nonlinear relationship between input parameters and output results. For solving the model parameters, the least squares method is used, and the calling process is relatively simple.

[0207] (2) Bubble Sort-GWO-ELM method proposed in the present invention

[0208] The model is built based on the extreme learning machine (ELM), which can capture nonlinear relationships, has faster training speed and better generalization ability, and is suitable for complex blasting prediction problems. The Grey Wolf Optimization (GWO) algorithm is used to automatically optimize model parameters, reducing manual intervention and improving the efficiency and accuracy of model parameter tuning.

[0209] 3. Model Validation

[0210] (1) Traditional method (MLR)

[0211] It usually uses simple cross-validation or leave-one-out method to validate the model, and the validation process is relatively simple. In addition, the interpretation of the validation results is more dependent on human experience and may be subjective.

[0212] (2) Bubble Sort-GWO-ELM method proposed in the present invention

[0213] It uses cross-validation combined with the GWO algorithm for multiple validations to ensure the accuracy and reliability of the validation results. And through a systematic validation process, it comprehensively evaluates the model performance.

[0214] 4. Example comparison

[0215] The present invention constructs a slope blasting data set based on data obtained from actual blasting projects. The data set includes data from multiple blasting operations performed in different time periods. During the construction of the data set, blasting data covering different geological conditions, blasting design schemes, and operating environments are selected as samples to ensure that the data is sufficiently representative to ensure the accuracy and reliability of the experimental results. After preprocessing and screening, the slope blasting data set is used to verify the multivariate linear regression (MLR) model and the Bubble Sort-GWO-ELM model proposed in the present invention. Table 1 shows the performance comparison results of the traditional MLR method and the BubbleSort-GWO-ELM model in predicting slope blasting block size.

[0216] Table 1

[0217]

[0218] It can be seen from Table 1 that, first, the prediction accuracy (R 2 ) is significantly higher than the traditional MLR method, indicating that this method can more accurately capture the changing trend of blasting block size; second, the mean absolute error (MAE) of this method is significantly lower than that of the MLR method, indicating that the deviation between the predicted value and the actual value of this method is smaller; third, the root mean square error (RMSE) of this method is much lower than that of the MLR method, indicating that the volatility and error margin of the prediction results of this method are smaller.

[0219] In summary, compared with the traditional multivariate linear regression method, the method based on the Bubble Sort-GWO-ELM model proposed in this invention has significant advantages in feature selection, model construction, model verification and practical application effect. Through scientific feature selection and parameter optimization, not only the prediction accuracy and efficiency are improved, but also the complex nonlinear relationship can be effectively dealt with, providing more accurate and reliable slope blasting block size prediction results.

[0220] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. The slope blasting fragmentation prediction method based on Bubble Sort-GWO-ELM model is characterized by: The following steps are involved: Step 1: Monitor blasting control parameter data at the construction site; Step 2: Preprocessing the monitored blasting control parameter data; Step 3: Screen the preprocessed blasting control parameter data, construct a blasting control parameter data set based on the screened blasting control parameter data, and divide the blasting control parameter data set into a training set and a test set; Step 4: Normalize the obtained training set and test set; Step 5: Based on the normalized training set, the parameters of the extreme learning machine ELM model are optimized using the improved gray wolf optimization algorithm to determine the optimal parameters of the optimized extreme learning machine ELM model; Step 6: Optimize the extreme learning machine ELM model with the optimal parameters, test the optimized extreme learning machine ELM model on the test set, count the output of the prediction of the test set samples, evaluate the optimal parameters, and establish a blasting effect prediction structure model based on the optimal parameters; Step 7: Use the blasting effect prediction model to predict the slope blasting fragmentation and output the prediction results.

2. The slope blasting fragmentation prediction method based on Bubble Sort-GWO-ELM model according to claim 1 is characterized in that: The specific steps of step 3 include: Step 3.1: Define the n monitored blasting control parameter data as an array A containing n elements. Each element in array A corresponds to a data to be screened and sorted. Each element X in array A is i It includes six types of blasting control parameters, namely tensile strength, blasting resistance line, step breaking angle, explosive consumption, hole spacing and row spacing; Step 3.2: Sort the n elements X in array A by using the bubble sort algorithm. i Sorting is performed, and based on the position of the elements in each round of sorting, different combinations of six types of blasting control parameters are achieved; Step 3.3: Calculate the squared difference of each of the different combinations according to the following formula: Where N is the number of data points in each combination, X i is the actual value monitored. is the mean of the data; Step 3.4: Repeat steps 3.2 and 3.3 until the elements in array A are arranged in ascending order, and then obtain the square differences of various combinations in each round of bubble sorting; Step 3.5: By comparing the square differences of various combinations, the blasting control parameter data corresponding to the combination with the smallest square difference value is selected; Step 3.6: Based on the blasting control parameter data screened in step 3.5, a blasting control parameter data set is established, and the blasting control parameter data set is divided into a training set and a test set.

3. The slope blasting fragmentation prediction method based on Bubble Sort-GWO-ELM model according to claim 2 is characterized in that: The specific steps of step 5 include: Step 5.1: Obtain the normalized training set data; Step 5.2: Initialize the number of hidden layer neurons L and activation function of the extreme learning machine ELM model; Step 5.3: Randomly initialize the input weight matrix W and the hidden layer bias b, and optimize the parameters W and b using the improved gray wolf optimization algorithm based on the data of the training set; Step 5.4: Based on the optimized parameters W and b, calculate the hidden layer output matrix H and determine the number of hidden layer neurons L; Step 5.5: Output the optimal parameters of the current extreme learning machine ELM model, including the number of hidden layer neurons L, W and b.

4. The slope blasting fragmentation prediction method based on Bubble Sort-GWO-ELM model according to claim 3 is characterized in that: The improved gray wolf optimization algorithm in step 5.3 includes the following steps: Step 5.3.1: Initialize the coefficient vectors A and C by: Where a is the convergence factor, which decreases linearly from 2 to 0 as the number of iterations t decreases, r1 and r2 are random vectors between [0, 1]; Step 5.3.2: Calculate the fitness value of each gray wolf in the wolf pack, mark the three wolf leaders with the largest fitness values ​​in the wolf pack as α, β, and δ, respectively, and mark the remaining gray wolves as ω. Calculate the position of the gray wolves using the following formula to surround the prey: X(t+1)=Ω·(X p (t)-A·D) D=|C·X p (t)-X(t)| Among them, t is the current iteration number, X p is the position vector of the current prey, X(t) is the position vector of the t-th generation leader, and Ω is the adaptive weight factor; and Step 5.3.3: Update the positions of the three alpha wolves in the wolf pack using the following formula: in, and Respectively represent the distances between α wolf, β wolf and δ wolf and other individuals; and Represent the current positions of α wolf, β wolf and δ wolf respectively; and is a random number, is the current location of the individual gray wolf; According to the positions of α, β and δ, the positions of all gray wolf individuals are updated by the following gray wolf individual position update formula: in, They represent the adjusted position of ω wolf affected by α wolf, β wolf and δ wolf respectively; in, Indicates the latest updated position of the gray wolf; Step 5.3.4: Determine whether the maximum number of iterations has been reached. If so, output the optimal solution. If not, return to step 5.3.

2.

5. The slope blasting fragmentation prediction method based on Bubble Sort-GWO-ELM model according to claim 4 is characterized in that: The specific steps of step 5.4 include: Step 5.4.1: For each input sample x i , the output h of the i-th hidden layer node is obtained according to the following formula i : h i =g(w i ,b i ,x)=g(w i x+b i ),w i ∈R D ,b i ∈R Among them, w i represents the input weight vector; b i represents the bias of the i-th neuron; x represents the input sample; R D represents D-dimensional real space; g(w i , b i , x) represents the activation function; Step 5.4.2: Based on the hidden layer output vectors of all samples, form the hidden layer output matrix H: H [h1(x)],...,h L (x)]; Where L represents the number of neurons in the hidden layer; Step 5.4.3: The output of the single hidden layer feedforward neural network ELM is obtained according to the following formula: Among them, β is the output weight between the hidden layer and the output layer; Step 5.4.4: Establish the optimization objective function of the output weight β based on the output weight β: min||Hβ-T|| 2 ,β∈R L×m Where H represents the output matrix of the hidden layer, T represents the target matrix, which contains the true output label or value of each training sample; m represents the number of training set samples; Step 5.4.5: Solve the above equation by minimizing the approximate square error method to obtain the optimal solution β * ; Step 5.4.6: According to the optimal solution β * Calculate the output matrix of the hidden layer.

6. A slope blasting fragmentation prediction system, characterized in that: The method is implemented based on any one of claims 1 to 5, comprising a data acquisition module, a data preprocessing module, a parameter screening module, a blasting effect prediction model establishment module and a slope blasting block size prediction module; Data acquisition module, used to monitor blasting control parameter data at the construction site; A data preprocessing module, used for preprocessing the data collected by the data acquisition module; A blasting control parameter data set construction module is used to filter the preprocessed blasting control parameter data by using bubble sort combined with square difference algorithm, construct a blasting control parameter data set based on the filtered blasting control parameter data, and divide the blasting control parameter data set into a training set and a test set; A normalization processing module is used to normalize the obtained training set and test set; A model optimization module is used to optimize the parameters of the extreme learning machine ELM model based on the normalized training set using the improved grey wolf optimization algorithm to determine the optimal parameters of the optimized extreme learning machine ELM model; The blasting effect prediction model establishment module is used to optimize the extreme learning machine ELM model with optimal parameters, test the optimized extreme learning machine ELM model on the test set, calculate the output of the prediction of the test set samples, evaluate the optimal parameters, and establish the blasting effect prediction model based on the optimal parameters; The slope blasting fragmentation prediction module is used to predict the slope blasting fragmentation using the blasting effect prediction model and output the prediction results.

7. A slope blasting fragmentation prediction system as claimed in claim 6, characterized in that: The model optimization module includes: A normalized data acquisition unit, used to acquire the data of the training set after normalization; A model initialization unit, used to initialize the number L of hidden layer neurons and activation function of the extreme learning machine ELM model; A parameter optimization unit is used to randomly initialize the input weight matrix W and the hidden layer bias b, and optimize the parameters W and b using the improved gray wolf optimization algorithm based on the data of the training set; A hidden layer neuron determination unit is used to calculate the hidden layer output matrix H and determine the number of hidden layer neurons L based on the optimized parameters W and b; The optimal parameter output unit is used to output the optimal parameters of the current extreme learning machine ELM model, including the number of hidden layer neurons L, W and b.

8. An electronic device, characterized in that: include: at least one processor; And, a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method as described in any one of claims 1 to 5.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images