A method, system and application for constructing a TSO-ELM model

By constructing the TSO-ELM model and optimizing the ELM network model using the TSO algorithm, the problems of low accuracy and low efficiency in the determination of rock block size after blasting were solved, achieving efficient and accurate block size prediction and optimizing blasting operations.

CN123115589APending Publication Date: 2026-11-17XI'AN UNIVERSITY OF ARCHITECTURE AND TECHNOLOGY +3
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411298255.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-18
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies suffer from low accuracy, low efficiency, and high cost in determining the size of rock blocks after blasting. Furthermore, traditional methods are greatly affected by human factors and cannot effectively control the block size.

Method used

The TSO-ELM model was adopted. The raw data of the blasted block size of the slope was preprocessed, and the ELM network model was optimized by using the TSO algorithm. The hidden layer weight values ​​were optimized by combining the spiral foraging strategy and the parabolic foraging strategy, and the TSO-ELM model was constructed for block size prediction.

Benefits of technology

It improves the accuracy and efficiency of blasting block size prediction, reduces the impact of human factors, and provides a scientific basis for optimizing blasting operations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115589A_ABST
    Figure CN123115589A_ABST
Patent Text Reader

Abstract

This invention provides a method, system, and application for constructing a TSO-ELM model, belonging to the field of slope blasting block size prediction technology. The method includes the following steps: acquiring raw data of slope blasting block size and preprocessing it to obtain preprocessed data; optimizing the ELM network model using the TSO algorithm to obtain the TSO-ELM model; inputting the preprocessed data into the TSO-ELM model for training and testing to obtain the optimized TSO-ELM model; and evaluating the prediction effect of the optimized TSO-ELM model on slope blasting block size using the output test results. This method utilizes the TSO algorithm to optimize the hidden layer weights in the ELM network model, and uses the unique spiral foraging strategy and parabolic foraging strategy of the TSO algorithm to quickly find the optimal solution for the factors affecting rock blasting block size. This enables a more accurate understanding and prediction of blasting block size distribution and influencing factors, providing a feasible research approach for blasting block size prediction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of slope blasting block size prediction technology, and in particular to a method, system and application of constructing a TSO-ELM model. Background Technology

[0002] In engineering blasting, the issue of rock block size refers to the uneven size (too large or too small) of the rock blocks after blasting, which affects subsequent construction and transportation. The issue of block size needs to be considered from the following aspects: 1. Selecting appropriate blasting parameters: Blasting parameters include explosive charge, throwing velocity, hole depth, etc. Selecting appropriate parameters can control the blasting effect, thereby controlling the block size; 2. Rationally arranging blasting holes: The location, shape, number, and density of blasting holes affect the blasting effect. Blasting holes should be rationally arranged according to specific circumstances to control block size; 3. Using segmented blasting technology: Segmented blasting refers to dividing large rocks into multiple segments for separate blasting to control block size. Before blasting, it is necessary to predict the hardness, toughness, fissures, and joints of the rock and formulate a reasonable segmentation plan; 4. Using pre-scraping technology: Before blasting, pre-cracks are created on the surface or inside the rock to promote block fracturing and reduce block size; 5. Controlling blasting vibration: Blasting vibration affects block size. Measures should be taken to control blasting vibration, such as reducing explosive charge, increasing hole spacing, and using buffer materials. Therefore, solving the block size problem in engineering blasting requires comprehensive consideration of multiple factors and the adoption of appropriate technologies and measures to achieve the goal of controlling block size.

[0003] Whether the quality of blasted block size control meets the control requirements is determined by the data from the measurement of large blocks after blasting. In practical traditional methods for measuring large blocks after blasting, there are three main approaches: First, the screening method, which uses specialized tools to screen and weigh the blasted ore and rock, and then uses mathematical calculations to extract the large block rate. This method has the advantage of high accuracy in practice, but also suffers from low efficiency and high cost. Second, the secondary blasting block size statistical method, which statistically analyzes the blasted ore and rock data within a defined area on-site to detect the blasting effect. This operation is relatively simple, but it cannot complete the overall measurement work. Third, the direct measurement method, which uses an assumed grid line to measure and calculate the blasted rock blocks, obtaining blasting data mathematically. This method provides intuitive data, but is significantly affected by human factors.

[0004] Nowadays, modern measurement methods using vision and machine learning cameras are commonly used in blasting measurements. With the support of computer technology, technicians complete the measurement work by inputting the captured and processed images into the computer system. The computer program processes the image signals, and digital signal conversion technology is used for blasting measurements, which improves both the quality and safety of the measurements.

[0005] Therefore, developing a machine learning-based method to predict factors influencing rock blasting block size has great potential and value in the field of blasting block size prediction. Summary of the Invention

[0006] In view of the deficiencies in the prior art, the present invention aims to provide a method, system and application for constructing a TSO-ELM model, so as to solve the problems existing in the background art.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] The first aspect of the technical solution disclosed in this invention is: a method for constructing a TSO-ELM model, comprising the following steps:

[0009] S1: Obtain the raw data of the blasting block size of the slope, and preprocess the raw data to obtain preprocessed data;

[0010] S2: Optimize the ELM network model using the TSO algorithm to obtain the TSO-ELM model;

[0011] S3: Input the preprocessed data into the TSO-ELM model for training and testing to obtain the optimized TSO-ELM model.

[0012] As a further preferred option of the above technical solution: In step S1, the original data is preprocessed using a correlation heatmap to obtain preprocessed data including average hole depth, minimum resistance line, explosive consumption, maximum single hole charge, and hole spacing.

[0013] As a further preferred embodiment of the above technical solution: In step S2, the step of optimizing the ELM network model using the TSO algorithm includes:

[0014] S21: Initialize the input layer weights and hidden layer biases of the ELM network model;

[0015] S22: The hidden layer weights of the ELM network model are optimized using the TSO algorithm, resulting in the TSO-ELM model.

[0016] Based on the above technical solution, further, in step S22, the step of optimizing the hidden layer weights of the ELM network model using the TSO algorithm includes:

[0017] S221: Initialize the parameters of the TSO algorithm and randomly generate parameter combinations;

[0018] S222: Optimize the hidden layer parameter weights of the ELM network model using the TSO algorithm, and use the RMSE of the ELM network model as the fitness function;

[0019] S223: Calculate the fitness value of individuals in the population and output the optimal solution;

[0020] S224: When the optimal individual cannot obtain the optimal solution, other individuals in the group update their positions and change the search range to find the global optimal solution;

[0021] S225: Determine whether the fitness value of the optimal population has reached the set threshold. If it has, terminate the optimization and obtain the optimal parameter combination; otherwise, return to step S221 to continue the optimization.

[0022] Furthermore, in step S224, the TSO algorithm updates the position of individuals in the population using two methods: a spiral foraging strategy and a parabolic foraging strategy.

[0023] Levy flight is used to improve the spiral foraging pattern of tuna. When the optimal individual in the tuna group fails to find food, the other tuna in the group update their positions according to the levy flight, expanding the individual's search range and finding the global optimum. The mathematical expression is as follows:

[0024]

[0025] in: Let be the position of the individual in the t-th generation selection; k1 and k2 are weighted coefficients controlling the individual's tendency to move towards the Levy flight individual and the previous individual. The values ​​of k1 and k2 are consistent with the values ​​of c1 and c2; D is the dimension of the position vector; u and v follow a normal distribution as shown in the following formula:

[0026]

[0027] Where λ takes the value λ = 1.5, and N is a normal distribution.

[0028] As a further preferred embodiment of the above technical solution: In step S4, the test results output by the optimized TSO-ELM model are compared with the custom D-BRR parameters to evaluate the prediction effect of the optimized TSO-ELM model on the blasting block size of the slope. The calculation formula for the D-BRR parameters is as follows:

[0029]

[0030] In the formula: bigrock refers to the large area after the blast, and smallrockarea refers to the small area after the blast.

[0031] Furthermore, the predictive performance of the optimized TSO-ELM model is evaluated using two indicators: MSE and MAE. The calculation formulas are as follows:

[0032]

[0033]

[0034] In the formula: Y i The measured value is f. i These are predicted values.

[0035] Another aspect of the technical solution disclosed in this invention is: a TSO-ELM model construction system, the system comprising:

[0036] The data acquisition module is used to obtain the raw data on the size of the blasted blocks on the slope;

[0037] The data preprocessing module is used to preprocess the raw data of the blasting block size of the slope to obtain preprocessed data;

[0038] The model building module uses the TSO algorithm to optimize the ELM network model, builds the TSO-ELM model, trains and tests the TSO-ELM model with preprocessed data, obtains the optimized TSO-ELM model, and deploys the optimized TSO-ELM model in the background.

[0039] The evaluation module uses the output test results to evaluate the prediction effect of the optimized TSO-ELM model on the blasting block size of slopes.

[0040] A further preferred embodiment of the above scheme is that the modeling module includes:

[0041] The initialization submodule is used to initialize the input layer weights and hidden layer biases of the ELM network model.

[0042] The parameter optimization submodule uses the TSO algorithm to optimize the hidden layer weights of the ELM network model. By optimizing the weights, the RMSE of the ELM network model is minimized, resulting in the TSO-ELM model.

[0043] Another aspect of the technical solution disclosed in this invention is: a method for predicting the blasting block size of slopes based on an optimized TSO-ELM model, comprising the following steps:

[0044] S1: Obtain the blasting parameters of the slope to be predicted and perform preprocessing;

[0045] S2: Input the preprocessed blasting parameters into the optimized TSO-ELM model and output the prediction results.

[0046] Compared with the prior art, the beneficial effects of the technical solution of the present invention are as follows:

[0047] 1. This invention combines the TSO algorithm with the ELM network model. The TSO algorithm is used to optimize the weights of the hidden layers in the ELM network model. The unique spiral foraging strategy and parabolic foraging strategy of the TSO algorithm are used to quickly find the optimal solution for the factors affecting the rock blasting block size, providing a new idea for the prediction of blasting block size.

[0048] 2. Compared with other existing models, the TSO-ELM model of this invention has advantages such as high prediction accuracy, fast convergence speed and high stability. It can more accurately understand and predict the block size distribution during the blasting process and the key factors affecting the blasting block size, providing a scientific basis for the optimization of blasting operations. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0050] Figure 1 This is a thermogram showing the correlation between the blasting parameters of the present invention.

[0051] Figure 2 The image recognition network structure of the UNet model of this invention;

[0052] Figure 3 This invention utilizes the UNet model to augment the original image set;

[0053] Figure 4 This invention utilizes the labelme tool to label images of rock blocks after blasting;

[0054] Figure 5 This is the neural network structure of the ELM network model of the present invention;

[0055] Figure 6 This is a flowchart illustrating the principle of the TSO-ELM model of the present invention;

[0056] Figure 7 This is a fitness curve of the TSO-ELM model of the present invention;

[0057] Figure 8 The fitness curve of the ELM network model of the present invention is shown.

[0058] Figure 9 The fitness curve of the SSA-ELM model of this invention is shown.

[0059] Figure 10 This is a comparison chart of the predicted and measured values ​​using the TSO-ELM model, ELM network model, and SSA-ELM model in this invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0061] Reference Figure 1-10 This invention proposes a method for constructing a TSO-ELM model, comprising the following steps:

[0062] A method for constructing a TSO-ELM model includes the following steps:

[0063] S1: Obtain the raw data of the blasting block size of the slope, and preprocess the raw data to obtain preprocessed data;

[0064] S2: Optimize the ELM network model using the TSO algorithm to obtain the TSO-ELM model;

[0065] S3: Input the preprocessed data into the TSO-ELM model for training and testing to obtain the optimized TSO-ELM model.

[0066] Specifically, in step S1, in order to study the influence relationship between the blasting block size and blasting parameters of highway slope, it is necessary to first obtain parameters suitable for the prediction model. In this invention, the characteristic value of foundation bearing capacity (F), average hole depth (D), minimum resistance line (L), explosive consumption (Q), maximum single hole charge (q), and hole spacing (l) are selected as network input layer parameters of the prediction model. The D-BRR parameter is a characterization parameter for quantifying the blasting block size defined by this invention. Some original data are shown in Table 1.

[0067] Table 1 Partial Original Dataset

[0068]

[0069]

[0070] To effectively identify the influence of blasting parameters on the D-BRR parameter (the pixel ratio of the blasting block size), this paper chooses to draw a parameter correlation heatmap. The heatmap method is a visualization technique used to display the correlations between variables in a data matrix. In statistics and data science, heatmaps are commonly used to display correlation coefficient matrices between variables, helping researchers quickly understand the strength and direction of the relationships between different variables. This method is particularly suitable for screening parameters that significantly affect the target variable in the initial data analysis stage.

[0071] The Pearson Correlation Coefficient (PCC) is commonly used in heatmaps to measure the linear relationship between variables. PCC is a statistical indicator describing the degree of linear correlation between two variables; its value ranges from -1 to 1, and its calculation formula is as follows:

[0072]

[0073] Where: x i and y i These are the observed values ​​of two variables; and It is the average of two variables; an r value of 1 indicates a perfect positive correlation, an r value of -1 indicates a perfect negative correlation, and an r value of 0 indicates no linear correlation. Therefore, the closer |r| is to 1, the greater the correlation.

[0074] Combination Figure 1 As shown, the correlation increases gradually from red to blue. Through detailed data exploration and visualization, the five parameters with the highest correlation were finally selected as the key input parameters of the model: average hole depth (DEEP), minimum resistance line (L), explosive consumption (Q), maximum single hole charge (q), and hole spacing (l). Retaining highly correlated parameters enhances the predictive ability of the model, simplifies the complexity of the model, and makes subsequent analysis and calculation more efficient and accurate.

[0075] In this invention, the original data of the blasted block size of the slope is obtained, and the original image of the blasted block size is preprocessed using the UNet model to obtain D-BRR parameter data.

[0076] The UNet model is a commonly used and simple segmentation model. Due to its efficiency and ease of construction, it is widely used by researchers. Its network structure, as shown in Figure 2, exhibits symmetry. In the UNet model, the encoder's feature maps are connected to the corresponding levels of the decoder via skip connections. This connection helps to transfer rich low-level and high-level feature information, contributing to improved segmentation accuracy. It is highly efficient due to its relatively few parameters and compact structure.

[0077] When preprocessing with the UNet model, the original image set of 113 photos was first augmented using methods such as brightness transformation, flipping, Gaussian blur, and salt-and-pepper noise, increasing the dataset by 17 times. Some results were observed. Figure 3 As shown.

[0078] Subsequently, the labelme tool was used to label the rock block images after blasting in detail, defining the boundaries and features of different rock blocks. The JSON file was then converted into images of the segmented regions. These labeled images were used as training data and input into the UNet model. Through network learning, the UNet model was able to identify and segment each rock block in the image. This method can effectively improve the accuracy and efficiency of rock block segmentation, providing technical support for subsequent applications such as rock mass analysis and fragmentation assessment.

[0079] like Figure 4 As shown, the burst areas are divided into large and small regions, and the formula for calculating the D-BRR parameter is:

[0080]

[0081] In the formula: bigrock refers to the large area after the blast, and smallrockarea refers to the small area after the blast.

[0082] The size distribution of burst blocks can often be approximated by a probability density function, a typical model of which is the Swebrec function, whose formula is as follows:

[0083]

[0084] Where: d is the particle size; d 60 This is the characteristic size, usually interpreted as ensuring that 50% of the rock blocks are smaller than this size; a and b are parameters for adjusting the curve shape. Assume d... 60 =60cm (assuming 50% of the material particles are less than or equal to 60cm), then the values ​​of a and b are a = 1.2 and b = 0.05 respectively.

[0085] by Figure 4 Using a basketball as a reference, the blasting block size of the rock is divided into large blocks (≥60cm) and small blocks (<60cm). Based on the Swebrec function and Kuz-Ram theory, calculations are performed. When the D-BRR parameter ≤29.570%, the blasting effect can be considered qualified. At the same time, the normalization constant is calculated to verify the result. The threshold size is 60.00cm when the normalization constant k = -3.322 and the D-BRR parameter = 29.57%.

[0086] In step 2, the Tuna Swarm Optimization (TSO) algorithm is combined with the Extreme Learning Machine (ELM) network model to analyze and predict the blasting block size of the slope and its influencing factors.

[0087] The TSO algorithm finds the optimal solution by simulating the foraging behavior of tuna schools. This algorithm is characterized by fast convergence and high accuracy in single-objective problems. Its inspiration comes from the foraging behavior of tuna schools, encompassing two strategies: 1) Spiral foraging: Tuna schools typically chase prey in a spiral formation, with adjacent tuna sharing information; 2) Parabolic foraging: Tuna schools typically form a parabolic formation around the prey, using it as a reference point. The metaheuristic optimization algorithm based on these two strategies and foraging populations is the TSO algorithm.

[0088] In the TSO algorithm, tuna schools chase prey by forming tight spirals. Besides chasing prey, the schools also exchange information. The parabolic foraging strategy involves tuna forming parabolic shapes with food as a reference point and searching for food in their surroundings. The mathematical expressions for the two strategies are as follows:

[0089] ①The expression for the spiral foraging strategy is:

[0090]

[0091]

[0092] β=e bl cos(2πb)

[0093]

[0094] in: Let be the position of the individual in the (t+1)th iteration; The current optimal individual position; c1 and c2 are multiple coefficients; t represents the current iteration number; t max This represents the maximum number of iterations; b is a random number in the range (0, 1).

[0095] When the optimal individual cannot find food, a coordinate is randomly generated in the search space as a reference point for spiral foraging.

[0096]

[0097] in, This represents a randomly generated coordinate point in the search space.

[0098] The expression for the parabolic foraging strategy is:

[0099]

[0100] Here, TF represents a random number with a value of 1 or -1.

[0101] By setting the selection probabilities of the two methods to be equal, the TSO algorithm updates the position of individuals in the population through parabolic foraging strategy and spiral foraging strategy throughout the optimization process, increasing the diversity of the search space, and continuously updating and calculating until the final condition is met.

[0102] In this invention, the TSO algorithm is mainly used to optimize the parameters of the ELM network model. The number of hidden layer nodes, activation function type, learning rate, and classification mode of the ELM network model are used as boundary parameters for the speed and position of the TSO algorithm. The specific process includes the following steps:

[0103] 1) Data normalization processing;

[0104] Since the selected factors affecting the blasting block size are different types of data with significant differences in their dimensions, the Min-MaxNormalization method is used to normalize these factors to eliminate the influence of different dimensions and ensure that the neural network of the ELM model can process the data more effectively.

[0105] The ELM network model is a single-hidden-layer feedforward neural network that uses randomized input layer weights and biases, and derives the output layer weights through generalized inverse matrix theory. Compared with traditional training models, the ELM network model has the advantages of simple structure and strong learning ability, and is widely used in fault diagnosis and effect prediction.

[0106] The network structure of the ELM network model is as follows: Figure 5 As shown, it consists of three layers: an input layer, a hidden layer, and an output layer, with each layer containing different neurons.

[0107] The N sets of blasting block size sample data collected along the slope will be represented as {x}. i ,t i}, the input training sample data is x in the input layer. i =[x i1 ,x i2 ,…,x in ] T ∈R n The output data of the training set target is t i =[t i1 ,t i2 ,…,t im ] T ∈R m Set the optimal training set to {x} i ,t i |x i ∈R D ,t i ∈R m,i=1,2,…,N;

[0108] After training the ELM network model, we can obtain:

[0109] Where, x i This represents a data example, t. i The labels represent the data examples, N represents the number of training samples; L is the number of hidden units; β j is the weight vector between the j-th hidden layer and the output; g is the activation function; b is the bias vector; x is the input vector;

[0110] Since the ELM network model approximates the training samples with minimal error, the target output of the training set is:

[0111]

[0112] Finally, the ELM network model matrix is ​​in the form of:

[0113] T = Hβ

[0114] Where, T∈R N×m It is the target output matrix of the training set; H∈R N×L Hidden layer output matrix; β∈R L×m Hidden layer outputs a weight matrix;

[0115] When the hidden layer is fed with weights and biases, H is a constant matrix, and the β solution is the least-squares particular solution of the linear system, i.e., the optimal solution sought by optimization:

[0116] Hβ * -T=min β ||Hβ-T||

[0117] Since H is invertible, the least squares solution β is... * Represented as:

[0118]

[0119] in: For H, the Moore-Penrose generalized inverse.

[0120] 2) Optimize the ELM network model using the TSO algorithm to construct the TSO-ELM model;

[0121] In this invention, the rock blocks produced by blasting are an important factor in evaluating the blasting effect. Their size, distribution and shape can all be studied. The factors affecting their properties mainly include geological conditions and blasting parameters.

[0122] Reference Figure 6Initialize the input layer weights and hidden layer biases of the ELM network model, and define the structure of the ELM neural network, including the number of input layer nodes and the number of hidden layer nodes. These parameters have a direct impact on the network's learning and generalization abilities.

[0123] The parameters of the TSO algorithm are initialized and random parameter combinations are generated. The hidden layer parameter weights of the ELM network model are optimized using the TSO algorithm, and the RMSE (root mean square error) of the ELM network model is used as the fitness function to form the TSO-ELM model. The goal is to minimize this error by optimizing the weights.

[0124] When training the TSO-ELM model, the parameters of the model are set as follows: the number of hidden layer nodes is set to 50, sigmoid or hard-limit is selected as the activation function according to the value of the variable, the learning rate is set to 0.01, the number of training iterations is 1000, and the population size is 30.

[0125] Velocity boundary conditions: The maximum velocity is set to [10, 0.1, 10, 100], and the minimum velocity is set to [-10, -0.1, -10, -100];

[0126] Location boundary conditions: The maximum location is set to [100,1,100,1000], and the minimum location is set to [1,0.01,10,100];

[0127] The normalized data was divided into training and testing sets in an 8:2 ratio. The training set was input into the TSO-ELM model, and the individual fitness values ​​were calculated and the optimal solution was output. Specifically, levy flight was used to improve the spiral foraging of tuna. When the best individual in the tuna group did not find food, the other tuna in the group updated their positions according to the levy flight, expanding the individual's search range and finding the global optimum. The fitting effect of the TSO-ELM model was optimized by modifying the parameter boundaries of parabolic foraging.

[0128] The mathematical expression for spiral foraging is as follows:

[0129]

[0130] In the formula: Let be the position of the individual in the t-th generation selection; k1 and k2 are weighted coefficients controlling the individual's tendency to move towards the Levy flight individual and the previous individual. The values ​​of k1 and k2 are consistent with the values ​​of c1 and c2; D is the dimension of the position vector; u and v follow a normal distribution as shown in the following formula:

[0131]

[0132] In the formula, λ takes the value λ = 1.5, and N is a normal distribution.

[0133] Determine whether the optimal population fitness value has reached the set threshold. If it has, terminate the optimization and obtain the optimal parameter combination; otherwise, return to continue the optimization.

[0134] After training the TSO-ELM model, the test set is used as input to the TSO-ELM model to output test results, resulting in the optimized TSO-ELM model. At the same time, the test results are used to evaluate the ability of the optimized TSO-ELM model to predict the size of blasting blocks, calculate the prediction error of the optimized TSO-ELM model on the test set, and statistically analyze the accuracy of the prediction results.

[0135] In the training and testing of the TSO-ELM model, this invention uses Matlab R2022b as the main computing and programming platform. Leveraging Matlab's powerful mathematical and visualization tools, it details how the TSO algorithm is combined to optimize the weights and bias parameters of the ELM network model. Furthermore, to evaluate the prediction accuracy of the optimized TSO-ELM model for blast block size, this invention employs two performance metrics: Mean Squared Error (MSE) and Mean Absolute Error (MAE). MAE assesses the deviation between predicted and actual values, while MSE assesses the distribution of errors. Their calculation formulas are as follows:

[0136]

[0137] In the formula: Y i The measured value is f. i These are predicted values.

[0138] The results of running the TSO-ELM model are shown in Table 2.

[0139] Table 2 Results of the TSO-ELM model

[0140]

[0141]

[0142] By deeply analyzing the performance of the TSO-ELM model on various datasets, this study further clarifies the mechanism and advantages of the TSO algorithm in optimizing ELM models, particularly highlighting the significant superiority of the TSO-ELM model in handling high-dimensional data and nonlinear problems. The research also includes visualization of the model training and prediction processes. Using Matlab's plotting capabilities, the study demonstrates the error descent and convergence during training, as well as the accuracy analysis of the prediction results. The experiment was conducted 50 times.

[0143] To comprehensively evaluate the effectiveness of the optimized TSO-ELM model, this invention selects the ELM network model and SSA-ELM (Sparrow Search Algorithm Fusion Extreme Learning Machine Model) as control groups, training and testing them with the same data. During the development of the optimized TSO-ELM model, the ELM network model was used as a control group. This comparative design allows for detailed analysis and verification of the specific impact of optimization measures on model performance. By comparing the performance of the three models on the same dataset, the superiority of the optimized TSO-ELM model over the ELM network model and the unoptimized model can be observed in various evaluation metrics, such as prediction accuracy, convergence speed, and model stability. Furthermore, the introduction of the SSA-ELM model as another optimized model for comparison further enhances the depth and breadth of the research.

[0144] The fitness curves, MAE, MSE, and prediction results of the optimized TSO-ELM model are compared with those of the other two models. The fitness comparison curves are shown in Figure 1. Figure 7-9 As shown in Table 5; the results of MSE and MAE are shown in Table 6; the curves comparing the prediction results of the three models with the measured data are shown in the figure. Figure 10 .

[0145] Table 5 Comparison of Predictive Performance Evaluation Results of the Three Models

[0146] Evaluation indicators ELM TSO-ELM optimization SSA-ELM MSE 2.133 0.437 11.685 MAE 1.116 0.811 3.670

[0147] Table 5 shows that different methods exhibit significant differences in performance across the mean squared error (MSE) and mean absolute error (MAE) metrics. Specifically, the optimized TSO-ELM model demonstrates significant improvements in both metrics, with its MSE decreasing to 0.437 and MAE to 0.811, indicating a substantial increase in prediction accuracy. In contrast, the ELM network model optimized using the SSA algorithm (i.e., SSA-ELM) performs the worst. In conclusion, the optimized TSO-ELM model exhibits the best prediction results, followed by the ELM network model, while SSA-ELM performs the worst.

[0148] like Figure 10The results show that, for the slope blasting block size dataset of this invention, the optimized TSO-ELM model exhibits faster and more stable sample regression fitting, and the predicted values ​​are closer to the measured data values. It significantly outperforms the traditional ELM network model in terms of accuracy, convergence speed, and stability. Furthermore, compared to the SSA-ELM model, the optimized TSO-ELM model also demonstrates superior performance on most performance metrics. This comparison verifies the significant effect of the TSO algorithm in improving the performance of ELM network models and confirms its applicability and superiority in complex data environments.

[0149] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a TSO-ELM model, characterized in that, Includes the following steps: S1: Obtain the raw data of the blasting block size of the slope, and preprocess the raw data to obtain preprocessed data; S2: Optimize the ELM network model using the TSO algorithm to obtain the TSO-ELM model; S3: Input the preprocessed data into the TSO-ELM model for training and testing to obtain the optimized TSO-ELM model.

2. The method for constructing a TSO-ELM model according to claim 1, characterized in that, In step S1, the raw data is preprocessed using a correlation heatmap to obtain the average hole depth, minimum resistance line, explosive consumption, maximum single-hole charge, and hole spacing.

3. The method for constructing a TSO-ELM model according to claim 1, characterized in that, Step S2, which involves optimizing the ELM network model using the TSO algorithm, includes the following steps: S21: Initialize the input layer weights and hidden layer biases of the ELM network model; S22: The hidden layer weights of the ELM network model are optimized using the TSO algorithm to obtain the TSO-ELM model.

4. The method for constructing a TSO-ELM model according to claim 3, characterized in that, Step S22, which involves optimizing the hidden layer weights of the ELM network model using the TSO algorithm, includes: S221: Initialize the parameters of the TSO algorithm and randomly generate parameter combinations; S222: Optimize the hidden layer parameter weights of the ELM network model using the TSO algorithm, and use the RMSE of the ELM network model as the fitness function; S223: Calculate the fitness value of individuals in the population and output the optimal solution; S224: When the optimal individual cannot obtain the optimal solution, other individuals in the group update their positions and change the search range to find the global optimal solution; S225: Determine whether the fitness value of the optimal population has reached the set threshold. If it has, terminate the optimization and obtain the optimal parameter combination; otherwise, return to step S221 to continue the optimization.

5. The method for constructing a TSO-ELM model according to claim 4, characterized in that, In step S224, the TSO algorithm updates the position of individuals in the population using two methods: spiral foraging strategy and parabolic foraging strategy. The steps for updating the position of individuals in the population using the spiral foraging strategy are as follows: Levy flight is used to improve the spiral foraging pattern of tuna. When the optimal individual in the tuna group fails to find food, the other tuna in the group update their positions according to the levy flight, expanding the individual's search range and finding the global optimum. The mathematical expression is as follows: in: Let be the position of the individual in the t-th generation selection; k1 and k2 are weighted coefficients controlling the individual's tendency to move towards the Levy flight individual and the previous individual, and the values ​​of k1 and k2 are consistent with the values ​​of c1 and c2; D is the dimension of the position vector; u and v follow a normal distribution as shown in the following formula: Where λ takes the value λ = 1.5, and N is a normal distribution.

6. The method for constructing a TSO-ELM model according to claim 5, characterized in that, In step S4, the test results of the optimized TSO-ELM model are compared with the D-BRR parameters to evaluate the prediction effect of the optimized TSO-ELM model on the blasting block size of the slope. The calculation formula for the D-BRR parameters is as follows: In the formula: bigrock refers to the large area after the blast, and smallrockarea refers to the small area after the blast.

7. A system for constructing a TSO-ELM model, characterized in that, The system is used to construct the method as described in any one of claims 1-6, the system comprising: The data acquisition module is used to obtain the raw data on the size of the blasted blocks on the slope; The data preprocessing module is used to preprocess the raw data of the blasting block size of the slope to obtain preprocessed data; The model building module uses the TSO algorithm to optimize the ELM network model, builds the TSO-ELM model, trains and tests the TSO-ELM model with preprocessed data, obtains the optimized TSO-ELM model, and deploys the optimized TSO-ELM model in the background. The evaluation module assesses the predictive performance of the optimized TSO-ELM model on the blasting block size of slopes.

8. The slope blasting block size prediction system based on the TSO-ELM model according to claim 7, characterized in that, The model building module includes: The initialization submodule is used to initialize the input layer weights and hidden layer biases of the ELM network model. The parameter optimization submodule uses the TSO algorithm to optimize the hidden layer weights of the ELM network model. By optimizing the weights, the RMSE of the ELM network model is minimized, resulting in the TSO-ELM model.

9. A method for predicting the blasting block size of slopes based on an optimized TSO-ELM model, characterized in that, Includes the following steps: S1: Obtain the blasting parameters of the slope to be predicted and perform preprocessing; S2: Input the preprocessed blasting parameters into the optimized TSO-ELM model obtained in claim 1, and output the prediction results of the blasting block size of the slope.