Kernel extreme learning machine-based blasting backrush distance intelligent prediction and control method

Abstract

The invention relates to an intelligent prediction and control method for a blasting back-rushing distance based on a nuclear extreme learning machine, which comprises the following steps of: collecting blasting design parameters and explosive parameters which influence the blasting back-rushing distance as input characteristics, and meanwhile, actually measuring the blasting back-rushing distance as an output characteristic; verifying and screening feature parameters in the original data to obtain final data, and constructing a training set and a test set; training a kernel extreme learning machine model based on the salat swarm optimization by using the training set to obtain a kernel extreme learning machine prediction model; performing performance test on the kernel extreme learning machine prediction model by using the test set; the method comprises the following steps: directly obtaining a predicted value of the blasting backshoot distance by inputting characteristic parameters into a kernel extreme learning machine prediction model, and finally, adjusting the input characteristic parameter values according to actual production requirements to control the blasting backshoot distance. According to the method, accurate prediction and effective control of the post-burst distance of strip mine blasting can be achieved, and the method has important practical value and scientific significance for reducing strip mine blasting harmfulness and ensuring the mining process.

Description

Technical Field

[0001] The invention belongs to the technical field of intelligent prediction and control of the post-blast throw distance in open-pit blasting of mining engineering, and specifically relates to an intelligent prediction and control method for post-blast throw distance based on a kernel extreme learning machine. Background Art

[0002] Blasting, as one of the efficient mining and excavation technical means for maintaining and increasing production capacity in open-pit mines, is still widely used in the mining operations of coal and other shallowly rich resources. However, the high proportion of dissipation of blasting energy results in only a small part of the explosive being used to break rocks and other geological bodies, and the propagation of the remaining high energy is difficult to control and extremely likely to cause post-blast hazards. Among them, the uncontrollable post-blast throw distance will have a great negative impact on subsequent blasting operations and the stability of rock and soil masses. For example, a post-blast throw distance of up to several meters in open-pit mine slope blasting will cause serious damage to the remaining surrounding rock, and it is bound to lead to the stagnation of subsequent blasting tasks due to the reduction of slope safety. Therefore, controlling the post-blast throw distance has important practical value and scientific significance for reducing open-pit mine blasting hazards and ensuring the mining process.

[0003] Accurate prediction of the post-blast throw distance is an important prerequisite for controlling and attempting to reduce its negative impact. At present, some technical means explore the post-blast throw mechanism in open-pit mines by constructing physical models, but the model consumables required for experiments and the test process are extremely time-consuming, labor-intensive, and costly, and the implementation process is relatively difficult. Some other technical means summarize the post-blast throw data generated after actual blasting operations, and then construct empirical formulas to predict the post-blast throw distance. Obviously, an empirical formula can mostly only be applicable to solving the problem of predicting the post-blast throw distance under specific mining conditions, and it is difficult to achieve universal application. With the improvement of computer software and hardware technologies, artificial intelligence algorithms have developed rapidly and have gradually been applied to the engineering field to solve various prediction and optimization problems. In recent years, there have been technologies that introduce artificial intelligence algorithms into traditional mining engineering technologies to attempt to solve problems such as blasting fragmentation, post-blast flying rock distance, and blasting vibration prediction, but there is still a lack of technical solutions that effectively combine artificial intelligence in the problem of post-blast throw prediction. However, how to use accurate prediction of post-blast throw to control the post-blast throw distance is still a technical problem faced by front-line mining industry workers. Ultimately, most post-blast throw problems are caused by adopting poor blasting design schemes, and designers lack technical means to real-time control the possible post-blast throw distances caused by different design schemes, thus unable to optimize the blasting design scheme. Therefore, there is an urgent need to provide an intelligent prediction and control method for post-blast throw distance based on a kernel extreme learning machine to help open-pit mines successfully complete mining tasks on the premise of controlling post-blast impacts. Summary of the Invention

[0004] Aiming at the problems existing in the above-mentioned prior art, the present invention provides an intelligent prediction and control method for the post-blasting throw distance based on a kernel extreme learning machine. The implementation process of this method is simple. It can accurately predict and effectively control the post-blasting throw distance of open-pit mines through artificial intelligence, and can provide theoretical and technical support for research such as optimizing blasting design, reducing blasting hazards, and improving mine production capacity. It has important practical value and scientific significance for reducing the harmfulness of open-pit mine blasting and ensuring the mining process, and can provide a reliable guarantee for the open-pit mine to successfully complete the mining task on the premise of controlling the post-blasting impact.

[0005] To achieve the above object, the present invention provides an intelligent prediction and control method for the post-blasting throw distance based on a kernel extreme learning machine, including the following steps:

[0006] Step 1: During the actual implementation of different blasting design schemes in an open-pit mine, record the blasting design parameters and explosive parameters that affect the post-blasting throw distance as input features. At the same time, actually measure the post-blasting throw distance after blasting operations and use it as an output feature. Take the input features and output features corresponding to each blasting operation as a set of original data samples. By collecting data during a large number of blasting operations, obtain several sets of original data samples and construct a database for the post-blasting throw distance.

[0007] Step 2: Inspect and screen all features in the database for the post-blasting throw distance. Use the screened features and the corresponding data samples to construct an original database, and then divide all the data in the original database into a training set and a test set according to a set ratio.

[0008] Step 3: Establish a kernel extreme learning machine model, and use the sand cat swarm optimization algorithm to optimize the hyperparameters of the kernel extreme learning machine model to obtain a kernel extreme learning machine model optimized by the sand cat swarm. Use the training set to train the kernel extreme learning machine model optimized by the sand cat swarm to obtain a kernel extreme learning machine prediction model.

[0009] Step 4: Use the test set to test the performance of the kernel extreme learning machine prediction model, and finally obtain a kernel extreme learning machine prediction model with excellent performance.

[0010] Step 5: Before actually implementing the blasting design plan, input the blasting design parameters and explosive parameters as input data into the kernel extreme learning machine prediction model. Use the kernel extreme learning machine prediction model to predict the post-blasting throw distance and output the predicted post-blasting throw distance. Compare the predicted post-blasting throw distance with the measured post-blasting throw distance, and adjust the blasting design parameters and explosive parameters according to the comparison result until the post-blasting throw distance predicted by the kernel extreme learning machine prediction model is consistent with the measured post-blasting throw distance, so as to obtain the blasting design parameters and explosive parameters required for actual production, and design a blasting plan with these blasting design parameters and explosive parameters, and control the post-blasting throw distance by implementing this blasting plan.

[0011] Further, in order to ensure that a variety of characteristic parameters can be comprehensively combined to accurately predict the post-blasting throw distance, in Step 1, the input features include hole length, hole spacing, burden, stemming length, unit explosive consumption, and drilling rate.

[0012] Further, in order to ensure the subsequent prediction accuracy, in Step 2, the process of checking and screening the features is as follows:

[0013] S21: Count all the input features obtained. Define each input feature as an element, and pair all elements in pairs as a group.

[0014] S22: According to formula (1), use the Kendall correlation coefficient calculation formula to obtain the Kendall correlation coefficient value Tau for each pair of elements in the group.

[0015]

[0016] In the formula, C represents the number of pairs of elements with consistent data samples, D represents the number of pairs of elements with inconsistent data samples, and N represents the total number of elements.

[0017] S23: Based on the obtained Kendall correlation coefficient value, check the correlation between the two features in this pair of elements in the group. If the Tau value is equal to 1, it indicates that there is a perfect positive correlation between the two features in this pair of elements; if the Tau value is equal to -1, it indicates that there is a perfect negative correlation between the two features in this pair of elements; if the Tau value is equal to 0, it indicates that there is no correlation between the two features in this pair of elements.

[0018] S24: Based on the test result of the correlation, delete the input features that have a high correlation with other input features and a low correlation with the output feature, and count the remaining qualified features and the corresponding sample data to form the final database.

[0019] As an optimization, in Step 2, all the data in the original database are divided into a training set and a test set according to a ratio of 7:3.

[0020] As an optimization, in step S3, the specific process of constructing a kernel extreme learning machine model optimized by the sand cat swarm is as follows:

[0021] S31: Establish the input layer of the kernel extreme learning machine model according to the selected input features, establish the hidden layer of the kernel extreme learning machine model according to the sample size corresponding to the input features, and establish the output layer of the kernel extreme learning machine model according to the output features;

[0022] S32: Introduce a kernel equation to improve the stability of the kernel extreme learning machine model, and obtain the predicted value of the kernel extreme learning machine model according to formula (2);

[0023]

[0024] In the formula, p i represents the i-th input feature, w represents the weight between the input layer and the hidden layer, b and G respectively represent the threshold and the activation matrix, u represents the weight between the hidden layer and the output layer, h and I respectively represent the number of neurons in the input layer and the hidden layer, R f represents the regularization parameter, I m represents the identity matrix, and K() represents the kernel equation;

[0025] S33: Use the sand cat swarm optimization algorithm to select the best hyperparameters of the kernel extreme learning machine model to obtain the kernel extreme learning machine model optimized by the sand cat swarm.

[0026] Furthermore, in order to obtain a kernel extreme learning machine model with excellent performance, in S33 of step S3, the specific process of constructing a kernel extreme learning machine model optimized by the sand cat swarm is as follows:

[0027] S33-1: Construct sand cat swarms with multiple set population sizes;

[0028] S33-2: Assign a set of independent hyperparameter candidate values of the kernel extreme learning machine model to each sand cat individual under a specific population size condition, and randomly distribute each sand cat individual in the predefined search space; complete the population initialization;

[0029] S33-3: Configure the hyperparameter combinations corresponding to each sand cat individual at different positions to the kernel extreme learning machine model to start iteration. Input the input features of the training set into the kernel extreme learning machine model to obtain the predicted values of the output features of the corresponding samples; calculate the error value between the predicted value and the measured value, and define this error value as the fitness, and then obtain the fitness value of each sand cat individual. At the same time, in each iteration process, make the sand cat individual continuously update its own position according to the set search strategy to find prey with lower fitness; after completing the set number of iterations, record the sand cat individual with the smallest fitness value in the current population size and its corresponding hyperparameter combination;

[0030] During the iteration process, the position P(t + 1) of the sand cat at the (t + 1)-th iteration is obtained according to formula (3).

[0031]

[0032] In the formula, P b (t) represents the best position of the sand cat at the t-th iteration, P c (t) represents the current position of the sand cat at the t-th iteration, r represents the sensitivity range of each sand cat, rand represents a random number, P rnd represents the random position during the process of the sand cat moving towards the food, cos(a) represents the moving direction of the sand cat, and R represents a factor that can influence whether the sand cat performs a search behavior or an attack behavior

[0033] S33 - 4: Change another specific population size condition, assign a set of independent hyperparameter candidate values of the kernel extreme learning machine model to each sand cat individual, and randomly distribute each sand cat individual within the predefined search space; complete the population initialization;

[0034] S33 - 5: Re - execute S33 - 3;

[0035] S33 - 6: Repeat S33 - 4 and S33 - 5 multiple times until the iteration process for multiple set population sizes is completed;

[0036] S33 - 7: Select the sand cat group with the minimum fitness value and relatively stable before the end of the iteration as the best population, take the sand cat individual with the best position within this population as the best individual, and take the hyperparameter combination corresponding to this best individual as the best hyperparameters of the kernel extreme learning machine model.

[0037] As an optimization, in S33 - 3 of step three, the number of iterations is set to 300.

[0038] As an optimization, in step four, the process of testing the performance of the kernel extreme learning prediction model using the test set is as follows:

[0039] S41: Input the input features in the test set into the kernel extreme learning machine prediction model to obtain the predicted values of the output features;

[0040] S42: Compare the measured values and predicted values of the output features of each sample, and calculate the performance evaluation indexes of the kernel extreme learning machine prediction model. The performance evaluation indexes include root mean square error RMSE, determination coefficient R 2 , mean absolute error MAE, and prediction quality U2;

[0041] S43: Initially determine the quality of the kernel extreme learning machine prediction model according to the obtained performance evaluation index values;

[0042] S44: Further compare the performance of the kernel extreme learning machine prediction model using a Taylor diagram and decide on the best prediction model.

[0043] As an optimization, in S42 of step four, calculate the root mean square error RMSE according to formula (4), calculate the determination coefficient R 2 according to formula (5), calculate the mean absolute error MAE according to formula (6), and calculate the prediction quality U2 according to formula (7);

[0044]

[0045] In the formula, n represents the number of samples in the test set; y m,i and y p,i respectively represent the measured value and predicted value of the output feature of the i-th sample in the test set, and averagey m,i represents the average value of all measured values of the output feature.

[0046] Furthermore, to facilitate the visualization of the operation process, in S44 of step four, after obtaining a kernel extreme learning machine prediction model with excellent performance, build a visualization prediction platform based on the kernel extreme learning machine prediction model and perform manual debugging. Based on the establishment of the visualization prediction platform, it is possible to conveniently adjust the input feature parameter values according to the actual production needs to control the post-blast throw distance; specifically, the operator only needs to input the blasting design parameters and explosive parameters on the platform to directly obtain the predicted value of the post-blast throw distance. Thus, the operator can quickly adjust the blasting design parameters and explosive parameters according to the required post-blast throw distance, and then can conveniently optimize the existing blasting design and control the post-blast throw distance, greatly facilitating the application of this method in actual engineering.

[0047] As an optimization, in S44 of step four, the process of building the visualization prediction platform is as follows:

[0048] S44-1: Design the main structure of the visualization prediction platform interface based on the MATLAB App designer software package;

[0049] S44-2: Embed the kernel extreme learning machine prediction model in the main structure of the visualization prediction platform interface, and make the input parameter part of the kernel extreme learning machine prediction model match and connect with the feature parameter input box on the visualization prediction platform interface, and make the prediction output result part of the kernel extreme learning machine prediction model match and connect with the prediction result display box on the visualization prediction platform interface;

[0050] S44-3: Manually debug the visualization prediction platform.

[0051] In the present invention, during the actual implementation of the blasting design plan, the blasting design parameters and explosive parameters that affect the post-blasting throw distance are used as input features, and the post-blasting throw distance is used as the output feature through actual measurement. Then, the corresponding input features and output features are used as a set of original data samples. Thus, it can be ensured that the original data samples contain the potential relationship between the input features and the output features, and this potential relationship is conducive to achieving accurate prediction of the post-blasting throw distance. By testing and screening all features, the features that are useless for predicting the post-blasting throw distance can be removed, and only the features that are useful for predicting the post-blasting throw distance are retained. In this way, it is beneficial to reduce the complexity of the prediction model. At the same time, it can effectively save computing power and is conducive to achieving an efficient prediction process. Establish a kernel extreme learning machine model, and use the sand cat swarm optimization algorithm to optimize the hyperparameters of the kernel extreme learning machine model, which can significantly improve the performance of the model and is conducive to achieving more accurate prediction of the post-blasting throw distance. Using the test set to test the performance of the kernel extreme learning machine prediction model is conducive to understanding the true performance of the model, and then a high-precision prediction model can be obtained, which can more effectively capture the potential relationship between the input features and the post-blasting throw, and can achieve more accurate prediction of the post-blasting throw distance.

[0052] The present invention can achieve accurate prediction of the post-blasting throw distance, and then can optimize the blasting design plan, and can effectively control the post-blasting throw distance based on the optimized method, which is beneficial to ensuring the reliable progress of subsequent blasting operations and is also beneficial to ensuring the stability of the rock and soil mass. The implementation process of this method is simple and has a high degree of intelligence. It realizes accurate prediction and effective control of the post-blasting throw distance in open-pit blasting through artificial intelligence, provides theoretical and technical support for research such as optimizing blasting design, reducing blasting hazards, and improving mine production capacity. It has important practical value and scientific significance for reducing the harmfulness of open-pit blasting and ensuring the mining process, and provides a reliable guarantee for the open-pit mine to successfully complete the mining task under the premise of controlling the post-blasting impact. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is the workflow diagram of the present invention;

[0054] Figure 2 is the feature correlation diagram in the present invention;

[0055] Figure 3 is the curve graph of the fitness of the prediction model in the present invention;

[0056] Figure 4 is the Taylor evaluation graph of the performance of the prediction model in the present invention;

[0057] Figure 5 This is a visualization program interface diagram for predicting and controlling the post-blast throw distance in the present invention. Specific embodiments

[0058] The present invention will be further described below with reference to the accompanying drawings.

[0059] As Figure 1 shown, the present invention provides an intelligent prediction and control method for post-blast throw distance based on a kernel extreme learning machine, including the following steps:

[0060] Step 1: During the actual implementation of different blasting design schemes in an open-pit mine, record the blasting design parameters and explosive parameters that affect the post-blast throw distance as input features. At the same time, actually measure the post-blast throw distance after blasting operations and use it as an output feature; take the input feature and output feature corresponding to each blasting operation as a set of original data samples, and obtain several sets of original data samples by collecting data during a large number of blasting operations to construct a post-blast throw distance database.

[0061] To ensure accurate prediction of the post-blast throw distance by comprehensively combining various characteristic parameters, the input features preferably include hole length, hole spacing, burden, stemming length, unit explosive consumption, and drilling rate in this embodiment.

[0062] Step 2: Check and screen all features in the post-blast throw distance database, use the screened features and corresponding data samples to construct an original database, and then divide all the data in the original database into a training set and a test set according to a set ratio; for the original database, divide the majority of the samples into the training set and the remaining minority samples are used to construct the test set. As a preference, all the data in the original database is divided into a training set and a test set according to a ratio of 7:3.

[0063] Among them, the statistical information of the input and output features in the training set and the test set are shown in Table 1 and Table 2 respectively;

[0064] Table 1: Statistical information table of input and output features in the training set

[0065]

[0066] Table 2: Statistical information table of input and output features in the test set

[0067]

[0068] Step 3: Establish a kernel extreme learning machine model, and use the sand cat swarm optimization algorithm to optimize the hyperparameters of the kernel extreme learning machine model to obtain a kernel extreme learning machine model based on sand cat swarm optimization; use the training set to train the kernel extreme learning machine model based on sand cat swarm optimization to obtain a kernel extreme learning machine prediction model.

[0069] Step 4: Use the test set to conduct performance inspection on the kernel extreme learning machine prediction model, and finally obtain a kernel extreme learning machine prediction model with excellent performance;

[0070] Step 5: Before actually implementing the blasting design plan, input the blasting design parameters and explosive parameters as input data into the kernel extreme learning machine prediction model, use the kernel extreme learning machine prediction model to predict the post-blasting throw distance, and output the predicted post-blasting throw distance; compare the predicted post-blasting throw distance with the measured post-blasting throw distance, and adjust the blasting design parameters and explosive parameters according to the comparison results until the predicted post-blasting throw distance by the kernel extreme learning machine prediction model is consistent with the measured post-blasting throw distance, obtain the blasting design parameters and explosive parameters required for actual production, and design a blasting plan with the blasting design parameters and explosive parameters, and control the post-blasting throw distance by implementing the blasting plan.

[0071] To ensure the subsequent prediction accuracy, in Step 2, the process of inspecting and screening the features is as follows:

[0072] S21: Statistically obtain all input features, define each input feature as an element, and pair all elements in pairs as a group;

[0073] S22: According to formula (1), use the Kendall correlation coefficient calculation formula to obtain the Kendall correlation coefficient value Tau for each group of element pairs;

[0074]

[0075] In the formula, C represents the number of pairs of elements with consistent data samples, D represents the number of pairs of elements with inconsistent data samples, and N represents the total number of elements;

[0076] S23: Based on the obtained Kendall correlation coefficient value, test the correlation between the two features in this group of element pairs. If the Tau value is equal to 1, it indicates that there is a perfect positive correlation between the two features in this group of element pairs. If the Tau value is equal to -1, it indicates that there is a perfect negative correlation between the two features in this group of element pairs. If the Tau value is equal to 0, it indicates that there is no correlation between the two features in this group of element pairs, as Figure 2 shown;

[0077] S24: Based on the test results of the correlation, delete the input features that have a high correlation with other input features and a low correlation with the output feature, and statistically obtain the remaining qualified features and the corresponding sample data to form the final database.

[0078] As an optimization, in Step 3, the specific process of constructing a kernel extreme learning machine model based on sand cat swarm optimization is as follows:

[0079] S31: Establish the input layer of the kernel extreme learning machine model based on the filtered input features (6 neurons can be set), establish the hidden layer of the kernel extreme learning machine model based on the sample size corresponding to the input features (164 neurons can be set), and establish the output layer of the kernel extreme learning machine model based on the output features (1 neuron can be set);

[0080] S32: Introduce a kernel equation to improve the stability of the kernel extreme learning machine model, and obtain the predicted value of the kernel extreme learning machine model according to formula (2);

[0081]

[0082] where p i represents the i-th input feature, w represents the weight between the input layer and the hidden layer, b and G respectively represent the threshold and the activation matrix, u represents the weight between the hidden layer and the output layer, h and I respectively represent the number of neurons in the input layer and the hidden layer, R f represents the regularization parameter, I m represents the identity matrix, and K() represents the kernel equation;

[0083] S33: Use the sand cat swarm optimization algorithm to select the best hyperparameters of the kernel extreme learning machine model, and obtain the kernel extreme learning machine model based on sand cat swarm optimization.

[0084] In order to obtain a kernel extreme learning machine model with excellent performance, in S33 of step three, the specific process of constructing the kernel extreme learning machine model based on sand cat swarm optimization is as follows:

[0085] S33-1: Construct sand cat swarms with multiple set population sizes; as an optimization, the population sizes are 25, 50, 100, and 200 respectively;

[0086] S33-2: Assign a set of independent hyperparameter candidate values of the kernel extreme learning machine model to each sand cat individual under a specific population size condition, and randomly distribute each sand cat individual within the predefined search space; complete the population initialization;

[0087] S33-3: Configure the hyperparameter combinations corresponding to each sand cat individual at different positions to the kernel extreme learning machine model to start iteration. Input the input features of the training set into the kernel extreme learning machine model to obtain the predicted values of the output features of the corresponding samples; calculate the error value between the predicted value and the measured value, and define this error value as the fitness, and then obtain the fitness value of each sand cat individual. At the same time, in each iteration process, make the sand cat individual continuously update its own position according to the set search strategy to find prey with lower fitness; after completing the set number of iterations, preferably set the number of iterations to 300, and record the sand cat individual with the minimum fitness value in the current population size and its corresponding hyperparameter combination;

[0088] During the iteration process, the position P(t + 1) of the sand cat at the (t + 1)-th iteration is obtained according to formula (3);

[0089]

[0090] In the formula, P b (t) represents the best position of the sand cat at the t-th iteration, P c (t) represents the current position of the sand cat at the t-th iteration, r represents the sensitivity range of each sand cat, rand represents a random number, P rnd represents the random position during the process of the sand cat moving towards the food, cos(a) represents the moving direction of the sand cat, and R represents a factor that can influence whether the sand cat performs a search behavior or an attack behavior

[0091] S33 - 4: Change another specific population size condition, assign a set of independent hyperparameter candidate values of the kernel extreme learning machine model to each sand cat individual, and randomly distribute each sand cat individual within the predefined search space; complete the population initialization;

[0092] S33 - 5: Re - execute S33 - 3;

[0093] S33 - 6: Repeat S33 - 4 and S33 - 5 multiple times until the iteration process for multiple set population sizes is completed;

[0094] S33 - 7: After the iteration is completed, output the fitness curve graph as Figure 3 shown. Select the sand cat group with the minimum fitness value (the minimum fitness value in this embodiment is 0.044161) and relatively stable before the end of the iteration (the population size in this embodiment is 50) as the best population, regard the sand cat individual with the best position within this population as the best individual, and regard the hyperparameter combination corresponding to this best individual as the best hyperparameters of the kernel extreme learning machine model.

[0095] As a preference, in step four, the process of using the test set to test the performance of the kernel extreme learning prediction model is as follows:

[0096] S41: Input the input features in the test set into the kernel extreme learning machine prediction model to obtain the predicted values of the output features;

[0097] S42: Compare the measured values and the predicted values of the output features of each sample, and calculate the performance evaluation indexes of the kernel extreme learning machine prediction model. The performance evaluation indexes include root mean square error RMSE, determination coefficient R 2 、mean absolute error MAE and prediction quality U2;

[0098] S43: Initially determine the performance quality of the kernel extreme learning machine prediction model based on the obtained performance evaluation index values, as shown in Table 3;

[0099] Table 3: Performance Evaluation Table of the Kernel Extreme Learning Machine Model

[0100]

[0101]

[0102] As an optimization, other machine learning models can be introduced simultaneously, such as artificial neural networks, support vector machines, and random forest to predict the post-blast throw distance, and the Taylor diagram as shown in Figure 4 is obtained to further compare the model performances and determine the best prediction model.

[0103] S44: Further compare the performances of the kernel extreme learning machine prediction model using the Taylor diagram and determine the best prediction model.

[0104] As an optimization, in S42 of Step Four, calculate the root mean square error RMSE according to formula (4), calculate the coefficient of determination R 2 , calculate the mean absolute error MAE according to formula (6), and calculate the prediction quality U2 according to formula (7);

[0105]

[0106] where n represents the number of samples in the test set; y m,i and y p,i represent the measured value and predicted value of the output feature of the i-th sample in the test set respectively, and averagey m,i represents the average value of all measured values of the output feature.

[0107] As shown in Figure 5 , to facilitate the visualization of the operation process, in S44 of Step Four, after obtaining a kernel extreme learning machine prediction model with excellent performance, build a visualization prediction platform based on the kernel extreme learning machine prediction model and conduct manual debugging. Based on the establishment of the visualization prediction platform, it is convenient to adjust the input feature parameter values according to the actual production needs to control the post-blast throw distance; specifically, the operator only needs to input the blasting design parameters and explosive parameters on the platform to directly obtain the predicted value of the post-blast throw distance. Thus, the operator can quickly adjust the blasting design parameters and explosive parameters according to the required post-blast throw distance, and then can conveniently optimize the existing blasting design and control the post-blast throw distance, which greatly facilitates the application of this method in actual engineering.

[0108] As an optimization, in S44 of Step 4, the process of building the visual prediction platform is as follows:

[0109] S44-1: Design the main structure of the visual prediction platform interface based on the MATLAB App designer software package. The main structure of the visual prediction platform interface includes a feature parameter input box, a result display box, a run button, a clear operation button, data export, etc.; specifically, the edit field command can be used to establish text or number boxes that can input each input feature, the button command can be used to arrange the run button and the clear operation button. The run button needs to add a callback function to call the feature numerical parameters in the text or number box for subsequent operations, and the statement 'cla(app.UIAxes)' needs to be added to the clear operation button to delete the previous operation to avoid interference between results. Then, the edit field command is used to establish text or number boxes that can display the output feature values, and the model prediction value is assigned to this command. Finally, the callback function of the 'export data' button is edited to export all predicted post-blasting throw distances and corresponding input feature values;

[0110] S44-2: Embed the kernel extreme learning machine prediction model in the main structure of the visual prediction platform interface, and match and connect the input parameter part of the kernel extreme learning machine prediction model (including hole length, hole spacing, burden, stemming length, unit explosive consumption, and drilling rate) with the feature parameter input box on the visual prediction platform interface, and match and connect the prediction output result part (post-blasting throw distance) of the kernel extreme learning machine prediction model with the prediction result display box on the visual prediction platform interface; then package all the code to generate the visual prediction platform interface as shown in Figure 5 the following figure;

[0111] S44-3: Manually debug the visual prediction platform. Specifically, multiple users (front-line mine operators and designers) can debug the visual prediction platform for 'one-key' operation according to the previous blasting design and the blasting design to be executed, including MATLAB starting the visual platform interface, manually inputting the input feature values corresponding to the blasting design, clicking the operation button to obtain the predicted post-blasting throw distance value using the developed prediction model, and simultaneously displaying the predicted value on the interface in real time. Click the clear button to perform the next prediction, and click the export button to achieve automatic data collection after all tests are completed.

[0112] In the present invention, during the actual implementation of the blasting design scheme, the blasting design parameters and explosive parameters that affect the post-blasting throw distance are used as input features, and the post-blasting throw distance is used as the output feature through actual measurement. Then, the corresponding input features and output features are used as a set of original data samples. Thus, it can be ensured that the original data samples contain the potential relationship between the input features and the output features, and this potential relationship is beneficial to the accurate prediction of the post-blasting throw distance. By testing and screening all features, the features that are useless for predicting the post-blasting throw distance can be removed, and only the features that are useful for predicting the post-blasting throw distance are retained. In this way, it is beneficial to reduce the complexity of the prediction model, and at the same time, it can effectively save computing power and is beneficial to realizing an efficient prediction process. Establishing a kernel extreme learning machine model and using the sand cat swarm optimization algorithm to optimize the hyperparameters of the kernel extreme learning machine model can significantly improve the performance of the model and is beneficial to realizing a more accurate prediction of the post-blasting throw distance. Using the test set to test the performance of the kernel extreme learning machine prediction model is beneficial to understanding the true performance of the model, and then a high-performance prediction model can be obtained, which can more effectively capture the potential relationship between the input features and the post-blasting throw and can realize a more accurate prediction of the post-blasting throw distance.

[0113] The present invention can realize the accurate prediction of the post-blasting throw distance, and then can realize the optimization of the blasting design scheme, and can effectively control the post-blasting throw distance based on the optimized method, which is beneficial to ensuring the reliable progress of subsequent blasting operations and is also beneficial to ensuring the stability of the rock and soil mass. The implementation process of this method is simple and has a high degree of intelligence. It realizes the accurate prediction and effective control of the post-blasting throw distance of open-pit mine blasting through artificial intelligence, provides theoretical and technical support for research such as optimizing blasting design, reducing blasting hazards, and improving mine production capacity. It has important practical value and scientific significance for reducing the harmfulness of open-pit mine blasting and ensuring the mining process, and provides a reliable guarantee for the open-pit mine to successfully complete the mining task on the premise of controlling the post-blasting impact.

Claims

1. An intelligent prediction and control method for the post-blasting backrush distance based on kernel extreme learning machine, characterized in that It includes the following steps: Step 1: During the actual implementation of different blasting design schemes in open-pit mines, record the blasting design parameters and explosive parameters that affect the throw distance after blasting as input features. At the same time, actually measure the throw distance after blasting operations and use it as the output feature; Take the input features and output features corresponding to each blasting operation as a set of original data samples. By collecting data during a large number of blasting operations, obtain several sets of original data samples and construct a throw distance database after blasting; Step 2: Check and screen all features in the throw distance database after blasting. Use the screened features and corresponding data samples to construct an original database, and then divide all the data in the original database into a training set and a test set according to a set ratio; Step 3: Establish a kernel extreme learning machine model, and use the sand cat swarm optimization algorithm to optimize the hyperparameters of the kernel extreme learning machine model to obtain a kernel extreme learning machine model based on sand cat swarm optimization; Use the training set to train the kernel extreme learning machine model based on sand cat swarm optimization to obtain a kernel extreme learning machine prediction model; Step 4: Use the test set to test the performance of the kernel extreme learning machine prediction model, and finally obtain a kernel extreme learning machine prediction model with excellent performance; Step 5: Before actually implementing the blasting design scheme, input the blasting design parameters and explosive parameters as input data into the kernel extreme learning machine prediction model, use the kernel extreme learning machine prediction model to predict the throw distance after blasting, and output the predicted throw distance after blasting; Compare the predicted throw distance after blasting with the measured throw distance after blasting, and adjust the blasting design parameters and explosive parameters according to the comparison results until the predicted throw distance after blasting by the kernel extreme learning machine prediction model is consistent with the measured throw distance after blasting, obtain the blasting design parameters and explosive parameters required for actual production, and design a blasting scheme with these blasting design parameters and explosive parameters, and control the throw distance after blasting by implementing this blasting scheme.

2. The intelligent prediction and control method for the post-blasting backrush distance based on the kernel extreme learning machine according to claim 1, characterized in that, In Step 1, the input features include hole length, hole spacing, burden, stemming length, unit explosive consumption, and drilling rate.

3. The intelligent prediction and control method for the post-blasting back-off distance based on the kernel extreme learning machine according to claim 2, wherein In Step 2, the process of checking and screening features is as follows: S21: Statistically obtain all input features, define each input feature as an element, and pair all elements in pairs as a group; S22: Use the Kendall correlation coefficient calculation formula to obtain the Kendall correlation coefficient value Tau for each pair of element pairings according to formula (1); In the formula, C represents the number of pairs of elements with consistent data samples, D represents the number of pairs of elements with inconsistent data samples, and N represents the total number of elements; S23: Based on the obtained Kendall correlation coefficient value, check the correlation between the two features for each pair of element pairings. If the Tau value is equal to 1, it indicates that there is a perfect positive correlation between the two features for this pair of element pairings. If the Tau value is equal to -1, it indicates that there is a perfect negative correlation between the two features for this pair of element pairings. If the Tau value is equal to 0, it indicates that there is no correlation between the two features for this pair of element pairings; S24: Based on the test results of correlation, delete the input features that have high correlation with other input features and low correlation with the output feature, and count the remaining qualified features and the corresponding sample data to form the final database.

4. An intelligent prediction and control method for the post-blasting backrush distance based on a kernel extreme learning machine according to claim 3, characterized in that In step two, divide all the data in the original database into a training set and a test set according to the ratio of 7:

3.

5. The intelligent prediction and control method for the post-blasting recoil distance based on the kernel extreme learning machine according to claim 4, characterized in that In step three, the specific process of constructing a kernel extreme learning machine model based on sand cat swarm optimization is as follows: S31: Establish the input layer of the kernel extreme learning machine model according to the filtered input features, establish the hidden layer of the kernel extreme learning machine model according to the sample size corresponding to the input features, and establish the output layer of the kernel extreme learning machine model according to the output feature; S32: Introduce a kernel equation to improve the stability of the kernel extreme learning machine model, and obtain the predicted value of the kernel extreme learning machine model according to formula (2); where p i represents the i-th input feature, w represents the weight between the input layer and the hidden layer, b and G represent the threshold and the activation matrix respectively, u represents the weight between the hidden layer and the output layer, h and I represent the number of neurons in the input layer and the hidden layer respectively, R f represents the regularization parameter, I m represents the identity matrix, and K() represents the kernel equation; S33: Use the sand cat swarm optimization algorithm to select the best hyperparameters of the kernel extreme learning machine model to obtain a kernel extreme learning machine model based on sand cat swarm optimization.

6. The intelligent prediction and control method for the post-blasting recoil distance based on the kernel extreme learning machine according to claim 4, characterized in that In S33 of step three, the specific process of constructing a kernel extreme learning machine model based on sand cat swarm optimization is as follows: S33-1: Construct sand cat swarms with various set population sizes; S33-2: Under a specific population size condition, assign a set of independent hyperparameter candidate values of the kernel extreme learning machine model to each sand cat individual, and randomly distribute each sand cat individual within the predefined search space; complete population initialization; S33-3: Configure the hyperparameter combination corresponding to each sand cat individual at different positions to the kernel extreme learning machine model to start iteration. Input the input features of the training set into the kernel extreme learning machine model to obtain the predicted value of the output feature of the corresponding sample; Calculate the error value between the predicted value and the measured value, and define this error value as the fitness, and then obtain the fitness value of each sand cat individual. At the same time, in each iteration process, make the sand cat individual continuously update its own position according to the set search strategy to search for prey with lower fitness; After completing the set number of iterations, record the sand cat individual with the minimum fitness value in the current population size and its corresponding hyperparameter combination; In the iteration process, obtain the position P(t + 1) of the sand cat at the (t + 1)-th iteration according to formula (3); Where, P b (t) represents the best position of the sand cat at the t-th iteration, P c (t) represents the current position of the sand cat at the t-th iteration, r represents the sensitivity range of each sand cat, rand represents a random number, P rnd represents the random position during the process of the sand cat moving towards the food, cos(a) represents the moving direction of the sand cat, and R represents a factor that can influence whether the sand cat performs a search behavior or an attack behavior S33-4: Change to another specific population size condition, assign a set of independent hyperparameter candidate values of the kernel extreme learning machine model to each sand cat individual, and randomly distribute each sand cat individual within the predefined search space; complete population initialization; S33-5: Re-execute S33-3; S33-6: Repeat S33-4 and S33-5 multiple times until the iteration process for various set population sizes is completed; S33-7: Select the sand cat swarm with the minimum fitness value and relatively stable before the end of iteration as the best population, regard the sand cat individual with the best position in this population as the best individual, and regard the hyperparameter combination corresponding to this best individual as the best hyperparameters of the kernel extreme learning machine model.

7. An intelligent prediction and control method for the post-blasting recoil distance based on a kernel extreme learning machine according to claim 5, characterized in that In S33-3 of step three, set the number of iterations to 300.

8. The intelligent prediction and control method for the post-blasting backrush distance based on the kernel extreme learning machine according to claim 6, characterized in that, In step four, the process of using the test set to test the performance of the kernel extreme learning prediction model is as follows: S41: Input the input features in the test set into the kernel extreme learning machine prediction model to obtain the predicted values of the output features; S42: Compare the measured values and predicted values of the output features of each sample, and calculate the performance evaluation indicators of the kernel extreme learning machine prediction model. The performance evaluation indicators include root mean square error RMSE, coefficient of determination R 2 , mean absolute error MAE, and prediction quality U2; S43: Initially determine the quality of the performance of the kernel extreme learning machine prediction model according to the obtained performance evaluation index values; S44: Further compare the performance of the kernel extreme learning machine prediction model using a Taylor diagram and determine the best prediction model.

9. The intelligent prediction and control method for the post-blasting backrush distance based on the kernel extreme learning machine according to claim 7, characterized in that, In S42 of Step 4, the root mean square error RMSE is calculated according to formula (4), and the coefficient of determination R is calculated according to formula (5). 2 The mean absolute error MAE is calculated according to formula (6), and the prediction quality U2 is calculated according to formula (7). where n represents the number of samples in the test set; y m,i and y p,i represent the measured value and the predicted value of the output feature of the i-th sample in the test set respectively, and averagey m,i represents the average value of all measured values of the output feature.

10. A method for intelligent prediction and control of post-blasting backrush distance based on kernel extreme learning machine according to claim 8, characterized in that, In S44 of Step Four, after obtaining a kernel extreme learning machine prediction model with excellent performance, build a visualization prediction platform based on the kernel extreme learning machine prediction model and perform manual debugging.

11. An intelligent prediction and control method for the post-blasting impact distance based on a kernel extreme learning machine according to claim 9, characterized in that, In S44 of Step Four, the process of building the visualization prediction platform is as follows: S44-1: Design the main structure of the visualization prediction platform interface based on the MATLAB App designer software package; S44-2: Embed the kernel extreme learning machine prediction model in the main structure of the visualization prediction platform interface, and make the input parameter part of the kernel extreme learning machine prediction model match and connect with the feature parameter input box on the visualization prediction platform interface, and make the prediction output result part of the kernel extreme learning machine prediction model match and connect with the prediction result display box on the visualization prediction platform interface; S44-3: Perform manual debugging on the visualization prediction platform.

Images