Blasting vibration prediction method and system based on integration algorithm

Abstract

The invention relates to a blasting vibration prediction method and system based on an integration algorithm, and belongs to the technical field of blasting engineering. The method comprises the following steps: obtaining a blasting vibration case database; correlation analysis is conducted on the blasting design parameters, the geological parameters and the blasting particle peak vibration speed, and prediction model input variables are determined; by calculating prediction results of a single gradient enhanced regression device, a single random forest regression device and a single decision tree regression device, integrating three base regression devices by adopting a voting algorithm, and generating a prediction model; optimizing hyper-parameters of the prediction model by using a genetic algorithm and nested 5-fold cross validation to obtain optimized hyper-parameters; a trained GA-VR model is obtained; according to the prediction model input variable, obtaining prediction sample data; and inputting the prediction sample data into the GA-VR model to obtain a prediction value. According to the method, high-precision blasting particle peak vibration velocity prediction can be realized, and the technical problem of errors caused by low precision of a traditional prediction model is solved.

Description

Technical Field

[0001] The invention relates to a blasting vibration prediction method and system based on an integrated algorithm, belonging to the technical field of blasting engineering. Background Art

[0002] Blasting technology is an engineering technology that uses the huge energy generated by explosives during the explosion process to deform, destroy and throw surrounding rocks and other original structures to achieve engineering purposes. Among them, the peak vibration velocity of the blasting particle is an important indicator to measure the impact of blasting vibration on the surrounding environment and structures. Accurately predicting the peak vibration velocity of the blasting particle is crucial to evaluating and controlling the impact of blasting vibration on surrounding buildings, infrastructure and personnel safety. Conventional machine learning models have low prediction accuracy. At the same time, due to the variability and uncertainty of the various influencing factors of the peak vibration velocity of the blasting particle, the impact on the calculation results is relatively large, and the peak vibration velocity of the blasting particle cannot be accurately predicted. Therefore, the low accuracy of conventional models in the prior art is the main technical obstacle to the prediction error of the peak vibration velocity of the blasting particle. Summary of the invention

[0003] The present invention provides a blasting vibration prediction method and system based on an integrated algorithm, which effectively solves the technical problem of errors caused by the low accuracy of traditional prediction models in the prior art. By integrating the advantages of a single model, high-precision prediction of the peak vibration velocity of blasting particles is achieved, providing a reference for reducing the impact of blasting vibration.

[0004] A blasting vibration prediction method based on an integrated algorithm, the method comprising: S1. Collect blasting vibration case data based on blasting design parameters, geological parameters and blasting particle peak vibration velocity, and build a historical blasting vibration case database; S2. Preprocessing the historical blasting vibration case database to obtain a blasting vibration case database; S3. Perform correlation analysis on the blasting design parameters, geological parameters and peak vibration velocity of the blasting particles to determine the input variables of the prediction model; S4. Using the blasting vibration case database to construct an integrated prediction model based on a voting algorithm, wherein the base regressor consists of a single gradient boosting regressor, a single random forest regressor, and a single decision tree regressor; S5. Set the prediction hyperparameter optimization target of the prediction model to maximize R 2 , optimizing the hyperparameters of the prediction model by combining genetic algorithm and nested 5-fold cross validation to obtain optimized hyperparameters; S6. Based on the optimization hyperparameters, the prediction model is trained according to the blasting vibration case database to obtain a trained GA-VR model; S7. According to the prediction model input variables, obtain prediction sample data; S8. Input the predicted sample data into the GA-VR model to obtain the predicted value , It is the peak vibration velocity of the explosion particle output by the GA-VR model.

[0005] The present invention also provides a blasting vibration prediction system based on an integrated algorithm, the system is used to execute a blasting vibration prediction method based on an integrated algorithm, the system comprising: A historical blasting vibration case database construction module, wherein the historical blasting vibration case database construction module is used to collect blasting vibration case data according to blasting design parameters, geological parameters and blasting particle peak vibration velocity; A blasting vibration case database acquisition module, wherein the blasting vibration case database acquisition module is used to perform data preprocessing on the historical blasting vibration case database to obtain a blasting vibration case database; A prediction model input variable determination module, the prediction model input variable determination module is used to perform a correlation analysis on the blasting design parameters, geological parameters and the peak vibration velocity of the blasting mass point to determine the prediction model input variables; A prediction model setting module, wherein the prediction model setting module integrates three base regressors, namely, a single gradient boosting regressor, a single random forest regressor, and a single decision tree regressor, using a voting algorithm to determine a prediction model; The hyperparameter optimization module is used to set the hyperparameter optimization target of the prediction model to maximize R 2 , optimizing the hyperparameters of the prediction model by combining genetic algorithm and nested 5-fold cross validation to obtain optimized hyperparameters; A GA-VR model acquisition module, wherein the GA-VR model acquisition module is used to train the prediction model based on the optimization hyperparameter and the blasting vibration case database to obtain a trained GA-VR model; A prediction sample data acquisition module, the prediction sample data acquisition module is used to obtain prediction sample data according to the prediction model input variables; A prediction value acquisition module is used to input the prediction sample data into the GA-VR model to obtain a prediction value, wherein the prediction value is , It is the peak vibration velocity of the explosion particle output by the GA-VR model.

[0006] The beneficial effects of the present invention are: The present invention realizes high-precision prediction of the peak vibration velocity of blasting particles by integrating the advantages of a single model, effectively solving the problem of low accuracy of blasting vibration prediction in the prior art, and comprehensively uses technical means such as data preprocessing, correlation analysis, integrated learning, and hyperparameter optimization to effectively quantify the prediction results of blasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 A schematic flow chart of a blasting vibration prediction method based on an integrated algorithm provided in an embodiment; Figure 2 It is a data sample correlation analysis diagram of the embodiment; Figure 3 This is a nested 5-fold cross validation diagram for hyperparameter optimization in the embodiment; Figure 4 It is a GA optimization iteration graph of an embodiment; Figure 5 A schematic diagram of the structure of a blasting vibration prediction system based on an integrated algorithm provided in an embodiment; Explanation of the accompanying drawings: historical blasting vibration case database construction module 1, blasting vibration case database acquisition module 2, prediction model input variable determination module 3, prediction model setting module 4, optimization hyperparameter acquisition module 5, GA-VR model acquisition module 6, prediction sample data acquisition module 7, prediction value acquisition module 8. DETAILED DESCRIPTION

[0008] The present invention provides a blasting vibration prediction method and system based on an integrated algorithm, which is used to solve the technical problem of errors in the prediction of the peak vibration velocity of blasting particles due to the low accuracy of traditional prediction models.

[0009] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0010] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0011] Example 1: Figure 1As shown, the present invention provides a blasting vibration prediction method based on an integrated algorithm, which is used to directly construct a prediction value of the peak vibration velocity of a blasting particle by setting different basis regressors and adopting a voting algorithm to integrate, and realizes high-precision prediction of the peak vibration velocity of a blasting particle by integrating the advantages of a single model, and provides effective information for decision makers to optimize blasting design parameters and select blasting geological conditions. The method includes: S1. Collect blasting vibration case data based on blasting design parameters, geological parameters and blasting particle peak vibration velocity, and construct a historical blasting vibration case database; wherein, the blasting design parameters are parameters used to describe specific blasting design and construction in blasting engineering, including maximum section charge, total charge, horizontal distance, height difference, front row resistance line, micro-difference time, pre-crack penetration rate and explosive detonation velocity; the geological parameters are parameters used to describe the physical and mechanical properties of rock or soil in blasting engineering, including the f value of the propagation medium, the integrity coefficient and the angle between the measuring point and the minimum resistance line direction; the blasting particle peak vibration velocity refers to the maximum velocity reached by particles in the surrounding medium (such as rock, soil or air) due to the shock wave and vibration wave generated by the explosion during the blasting process, which is used to measure the impact of blasting vibration; by adopting existing research results data, and / or collecting blasting design parameters, geological parameters and blasting particle peak vibration velocity data at the actual site, a historical blasting vibration case database is constructed.

[0012] S2. After constructing the historical blasting vibration case database, data preprocessing is performed on the historical blasting vibration case database to obtain a blasting vibration case database; the data preprocessing includes data cleaning, data conversion, feature selection, data grouping, data formatting, etc.

[0013] S3. Carry out correlation analysis on the blasting design parameters, geological parameters and the peak vibration velocity of the blasting particles. Analyze the correlation between the blasting design parameters, geological parameters and the peak vibration velocity of the blasting particles by calculating feature importance, correlation coefficient, chi-square test, regression analysis and other methods. Through correlation analysis, screen out the factors that have a greater impact on the blasting vibration. According to the analysis results, use the parameters with a higher correlation with the blasting vibration as input variables for determining the prediction model.

[0014] S4. Use the blasting vibration case database to build a prediction model based on the voting algorithm for integration, where the base regressor consists of a single gradient boosting regressor, a single random forest regressor, and a single decision tree regressor; the single gradient boosting regressor is an accurate, flexible, and highly interpretable integrated learning algorithm, the single random forest regressor is an accurate, robust, and easy-to-implement multi-tree integrated prediction framework, the single decision tree regressor is an intuitive, easy-to-understand, feature-selection-based prediction model, and the voting regression integrated model is an integrated learning framework that integrates multiple prediction advantages, improves generalization capabilities, and enhances model robustness, and is used to process large-scale data and is applied to a variety of machine learning tasks. Import the NumPy and Pandas libraries for subsequent data processing and model training; specifically, including: S41. The base regressors trained on the same new data set are integrated using a voting algorithm to obtain an integrated model; the base regressors include a single gradient boosting regressor, a single random forest regressor, and a single decision tree regressor; S42. Divide the input data set, and divide the prediction model input variables into training sets T train and the test set T test , the corresponding outputs are recorded as P train and P test ; S43. Train the base regressor using the dataset T train and P train The base regressors are trained, and the base regressors include a single gradient boosting regressor, a single random forest regressor, and a single decision tree regressor. The trained base regressors are used to predict the test set T test , and its predicted output is recorded as P t1 , P t2 and P t3 ; S44. The key hyperparameters in the voting ensemble model include: the number of decision trees constructed in a single gradient boosting regressor n_estimators, the maximum depth of the decision tree max_depth, the learning rate learning_rate, and the minimum number of samples required for node splitting min_samples_split; the maximum depth of the decision tree max_depth, the number of decision trees constructed n_estimators, and the minimum number of samples required for node splitting min_samples_split in a single random forest regressor; the minimum number of samples required for node splitting min_samples_split in a single decision tree regressor, the maximum number of leaf nodes in the decision tree max_leaf_nodes, the maximum depth of the decision tree max_depth, and the proportion of the three base regressors of the voting algorithm.

[0015] S5. Set the prediction hyperparameter optimization target of the prediction model to maximize R 2 , hyperparameters include learning rate, maximum depth of tree, minimum number of samples for splitting, etc. In order to evaluate the performance of each hyperparameter combination, a fitness function is constructed using R 2 The indicator is used as the fitness function, and the hyperparameter optimization goal is to maximize R 2 It means using the median regression model to set different hyperparameters to iterate the maximum R 2 ,in , is the predicted value of the prediction model, is the data sample label, is the average value of the data sample labels, that is, the average value of the true observations; The hyperparameters of the prediction model are optimized by combining genetic algorithm and nested 5-fold cross validation to obtain the optimized hyperparameters. Genetic algorithm is a highly parallel, random, adaptive global optimization probability search algorithm developed by referring to the natural selection and evolution mechanism of the biological world. It simulates the phenomena of selection, crossover and mutation in natural selection and inheritance. It first generates an initial population, and through selection, crossover and mutation operations, generates a group of individuals that are more suitable for the environment. After evolution from generation to generation, the population finally converges to a group of individuals that are most suitable for the environment, and the optimal solution to the problem is obtained. The genetic algorithm optimized voting ensemble model has a global search capability when searching the hyperparameter space, which helps to find a better model configuration, thereby improving the performance and generalization ability of the GA-VR blasting vibration prediction model, so as to obtain better prediction accuracy and effect in practical applications. First, the population is initialized using a random method to generate a set of random hyperparameter combinations, and then the genetic algorithm is used to optimize the population. Through operations such as selection, crossover and mutation, the population is continuously iterated to gradually improve the value of the fitness function. In each iteration, each hyperparameter combination in the population is brought into the nested 5-fold cross validation (see Figure 3 ), in each cross-validation, each hyperparameter combination is evaluated and selected according to the value of the fitness function, and then according to the results of the nested 5-fold cross-validation, the hyperparameter combination with higher fitness is selected as the new population, and the population is updated. Finally, according to the preset termination condition (such as reaching the maximum number of iterations or the fitness reaching the preset threshold), the iteration is stopped and the optimal hyperparameter combination is output.

[0016] S6. Based on the optimization of the hyperparameters, the prediction model is trained according to the blasting vibration case database, and the blasting vibration case database is loaded from the storage system to train the voting integrated model. During the training process, the genetic algorithm is used to optimize the hyperparameters, and the nested 5-fold cross validation is used to evaluate the performance of the model. After the training is completed, the GA-VR model (an integrated learning algorithm that combines the genetic algorithm and the voting integrated model) is obtained. The GA-VR model is evaluated using the nested 5-fold cross validation to calculate the predicted R 2 The performance of the model is judged by using evaluation indicators such as value. According to the evaluation results, the GA-VR model is optimized, such as adjusting hyperparameters and increasing data samples, to improve the performance of the model.

[0017] S7. Obtain prediction sample data according to the prediction model input variables; that is, prepare corresponding prediction sample data according to the input variables determined above, and these prediction sample data are data after preprocessing and assignment.

[0018] S8. Input the predicted sample data into the GA-VR model. In the GA-VR model, the input variables are used as the input of the model. The trained model makes predictions based on these input variables, and the prediction results are output in the form of predicted values, where the predicted values ​​are , The predicted value of the peak vibration velocity of the blasting particle output by the GA-VR model is achieved by integrating the advantages of a single model to achieve high-precision prediction of the peak vibration velocity of the blasting particle, solving the technical problem of errors caused by the low accuracy of the traditional prediction model.

[0019] In a preferred implementation manner provided by an embodiment of the present invention, preprocessing the historical blasting vibration case database includes: Multidimensional preprocessing indicators are obtained, wherein the multidimensional preprocessing indicators include duplicate data elimination, missing data elimination and abnormal data elimination. Duplicate data refers to data samples that are repeated due to repeated collection; missing data refers to the presence of parameters in the data sample that cannot be obtained on site or are statistically omitted; abnormal data refers to the presence of parameters in the data sample due to subjective statistical errors or objective abnormalities in on-site statistics, which are generally data that cannot exist based on experience, including non-positive maximum segment charge or non-positive peak vibration velocity of blasting particles, etc.

[0020] The method to eliminate duplicate data is to determine the unique identifier of each data record, such as timestamp, ID, etc., use the determined identifier to detect the data in the database, find duplicate data records, delete or merge duplicate data records, and ensure that the database contains only unique data records. The method to eliminate missing data is to check each data record in the database, find the location of the missing data, and select a method to fill the missing data according to the actual situation, such as using statistical methods such as mean, median, mode, or using the previous or next valid data to fill. If the missing data cannot be effectively filled, the entire data record containing the missing data will be deleted. The method to eliminate abnormal data is to formulate the standard and range for judging abnormal data based on the characteristics and actual situation of the data. For example, according to the 3 statistical methods. The judgment criteria for abnormal data are determined based on principles or empirical formulas, and the determined judgment criteria are used to detect the data in the database, find abnormal data records, and delete the abnormal data records to ensure that the database contains only valid and reliable data records. After the steps of removing duplicate data, missing data, and abnormal data, a blasting vibration case database is generated. The database is shown in Table 1, where blasting design parameters and geological parameters are data sample features, and the peak vibration velocity of blasting particles is the data sample label. 80% of the data in the database is divided into a training set for the prediction model, and 20% is divided into a test set for the prediction model. This preferred implementation method makes data analysis more accurate and reliable by removing duplicate data, filling missing data, and correcting abnormal data, thereby achieving the technical effect of improving data quality.

[0021] Table 1 Blasting vibration case database ; In another preferred implementation manner provided by the embodiment of the present invention, correlation analysis is performed on the blasting design parameters, geological parameters and the peak vibration velocity of the blasting mass point to determine the input variables of the prediction model, including: Calculate the Pearson correlation coefficient between each blasting design parameter and geological parameter. The Pearson correlation coefficient is a statistic that measures the degree of linear correlation between two variables. The Pearson correlation coefficient calculation formula is: , where E(·) represents expectation, A and B represent the characteristics of two different sets of blasting design parameters and geological parameters of the data sample, the positive and negative Pearson correlation coefficients represent positive and negative correlations, respectively, and the absolute value represents the magnitude of the correlation. Indicates that there is no correlation between features. Indicates that there is a weak correlation between features. It indicates that the correlation between features is good. Indicates that there is a strong correlation between features.

[0022] Statistics are collected on the distribution of blasting design parameters and geological parameter attributes in the blasting vibration case database, and the distribution of blasting design parameters and geological parameter attributes (features and labels) in the statistical data set. Checking the independence of attributes means collecting statistics on each attribute (features and labels) in the data set to understand the value range, frequency distribution, missing values, etc. of each attribute, and evaluating whether there is a correlation or dependency between them. Statistics are collected on the distribution of sample data labels that are not less than and less than the dividing point in each blasting design parameter and geological parameter feature.

[0023] As shown in Table 2, the extreme gradient boosting algorithm (ExtremeGradient Boosting, XGBoost) is used to sort the characteristic importance of the blasting design parameters and geological parameters, and then combined with the correlation coefficient Table 3, it can be seen that for the blasting design parameters, there is a strong correlation between the horizontal distance and the height difference, and the characteristic importance of the height difference is low; for the geological parameters, there is a strong correlation between the integrity coefficient and the f value of the propagation medium, and for the geological parameters, there is a weak correlation between the front resistance line and the angle between the measuring point and the minimum resistance line. There is no multicollinearity effect between the other parameters.

[0024] Table 2 Feature importance ranking ; Table 3 Correlation coefficients ; After removing the height difference and integrity coefficient, a kernel density estimation map is established, such as Figure 2 As shown in the figure, the kernel density estimation diagram in the diagonal part is the density estimation of the corresponding attribute. According to the upper triangular part of the matrix, the correlation between the nine features can be more intuitively obtained through the correlation coefficient R. From the correlation coefficient R, it can be seen that the feature correlations of the data after dimensionality reduction are all weak correlations. Therefore, the kernel density estimation diagram is combined to delete the height difference and integrity coefficient features to reduce the dimensionality of data sample features.

[0025] Selecting the prediction model input variables according to the Pearson correlation coefficient and the blasting design parameters and geological parameter attribute distribution means selecting key factors closely related to blasting vibration as the prediction model input variables by comprehensively considering the correlation between each blasting design parameter, geological parameter and the peak vibration velocity of the blasting particle, as well as the value range and distribution characteristics of each parameter. This preferred implementation method comprehensively and accurately understands the data distribution and correlation, and provides easy-to-understand visualization results, achieving the technical effect of improving the quality and reliability of the prediction model.

[0026] In another preferred implementation manner provided by an embodiment of the present invention, training the prediction model according to the blasting vibration case database includes: When the training is not completed, the hyperparameters of the prediction model are selected, crossed, and mutated according to the genetic algorithm and nested 5-fold cross validation to obtain the hyperparameters of the updated prediction model. The selection refers to selecting a part of the solutions from the current set of candidate solutions for generating the next generation, using an evaluation function to measure the quality of the solutions, and performing probabilistic selection based on the quality of the solutions to ensure that better solutions have a higher chance of being selected; the crossover refers to creating a new offspring solution from two or more parent solutions, combining part of the information of different parent solutions through a crossover operation to generate a new solution with some parent characteristics; the mutation refers to introducing some randomness or small changes in the new generation of solutions to maintain diversity, avoid falling into local optimal solutions, and have the opportunity to discover the global optimal solution. When the training process of the model does not meet the evaluation indicators, the selection, crossover, and mutation operations in the genetic algorithm are used to update the hyperparameters of the prediction model to obtain better prediction performance. In this preferred embodiment, the evaluation indicators are selected: R 2 and MSE. Among them: ; The hyperparameters of the prediction model are iteratively optimized to obtain iterative optimization hyperparameters. In this preferred embodiment, there are 10 groups of individuals in the initial population of the genetic algorithm, corresponding to 10 hyperparameters of the three base regressors (n_estimators, max_depth, learning_rate and min_samples_split in a single gradient enhancement regressor; max_depth, n_estimators and min_samples_split in a single random forest regressor; min_samples_split, max_leaf_nodes and max_depth in a single decision tree regressor) and 3 hyperparameters of the integrated model (the proportion of the three base regressor models). Among them, max_depth specifies the maximum depth of each decision tree, defines the maximum number of split layers or node depth allowed in the tree, and min_samples_split refers to the minimum number of samples required when performing node splitting. In the single gradient boosting regressor model, n_estimators specifies the number of decision trees built during the training process, and learning_rate is the learning rate, which controls the adjustment amplitude of the model weight in each iteration, determines the contribution of each tree and the size of the weight update learning rate; in the single random forest regressor model, n_estimators refers to the number of decision trees to be built, and this parameter directly affects the complexity and performance of the model; in the single decision tree regressor model, max_leaf_nodes specifies the maximum number of leaf nodes in the decision tree. By limiting the number of leaf nodes, the complexity of the decision tree can be controlled, thereby affecting the overfitting and generalization ability of the model. Set the initial population individual value ranges corresponding to max_depth, min_samples_split, n_estimators and learning_rate to (3, 10), (5, 12), (75, 105) and (0.020, 0.030) respectively; set the initial population individual value ranges corresponding to max_depth, n_estimators and min_samples_split to (5, 20), (150, 300) and (2, 10) respectively; set the initial population individual value ranges corresponding to max_depth, max_leaf_nodes and min_samples_split to (2, 30), (10, 50) and (2, 20) respectively.

[0027] In this preferred embodiment, Figure 3As shown, the two cycles of nested 5-fold cross-validation respectively randomly divide the prediction model training set into 5 parts, namely SET1, SET2, SET3, SET4 and SET5. Four of them are taken as nested cross-validation training sets (such as Train1, Train2, Train3 and Train4) for training to obtain the uncertainty prediction model of blasting vibration, and the remaining one is used as the nested cross-validation test set (such as Test5) for prediction result evaluation, which is carried out 5 times in total. The average value of the evaluation results of the 5 inner cycles is taken as the optimal hyperparameter value, and the average value of the evaluation results of the 5 outer cycles with this hyperparameter is taken as the mean of the prediction evaluation index of this hyperparameter. Among them, R 2 and MSE values ​​are used to evaluate the prediction effect, R 2 The closer it is to 1, the closer MSE is to 0, indicating that the model prediction is more accurate. Figure 4 (a) shows the process of using genetic algorithm to optimize a single gradient boosting regressor for prediction. When the genetic algorithm iterates to step 176, the optimal R 2 score (= 0.941), yielding the best MSE score (= 0.1136); Figure 4 (b) shows the process of using genetic algorithm to optimize a single random forest regressor for prediction. When the genetic algorithm iterates to step 39, the optimal R 2 score (= 0.932), producing the best MSE score (= 0.1314); Figure 4 (c) shows the process of using genetic optimization to predict a single decision tree regressor. When the genetic algorithm iterates to step 3, the optimal R 2 score (= 0.879), yielding the best MSE score (= 0.2351); Figure 4 (d) shows the process of using genetic algorithm to optimize voting algorithm for prediction. When the genetic algorithm iterates to step 79, the optimal R 2 score (= 0.949), yielding the best MSE score (= 0.0992).

[0028] Based on the iterative optimization of hyperparameters, training is continued according to the updated blasting vibration case database. In this preferred embodiment, after iterative optimization, the optimal hyperparameters for prediction of the GA-GBR model, GA-RFR model and GA-DTR model are shown in Table 4, and the optimal hyperparameters for the integrated model GA-VR model are shown in Table 5. The model is continued to be trained and optimized using the optimized hyperparameter configuration and the updated blasting vibration case database to further improve the performance and accuracy of the model and obtain better prediction results. This preferred embodiment finds a set of optimal hyperparameter configurations through repeated training and optimization, so that the model can maintain good performance and accuracy under various conditions, achieving the technical effect of improving the stability and robustness of the model.

[0029] Table 4 Optimal hyperparameters for prediction of GA-GBR model, GA-RFR model and GA-DTR model ; Table 5 Optimal hyperparameters of the integrated model GA-VR model ; In another preferred implementation provided by an embodiment of the present invention, the nested 5-fold cross validation includes an inner loop and an outer loop, the inner loop searches for optimal hyperparameters based on genetic algorithm optimization, and provides optimal hyperparameters to the outer loop, the outer loop provides data samples for training to the inner loop, and at the same time, the outer layer reserves some data samples as a test set to test the effect of the inner loop model. This preferred implementation adopts nested 5-fold cross validation to prevent overfitting, and searches for hyperparameters through model generalization error estimation, achieving the technical effect of enabling the model to obtain the optimal parameters.

[0030] Embodiment 2: Based on the same inventive concept as the blasting vibration prediction method and system based on integrated algorithm in the above embodiment, Figure 5 As shown, the present invention provides a blasting vibration prediction system based on an integrated algorithm, and the system and method embodiments in the embodiments of the present invention are based on the same inventive concept. The system includes: A historical blasting vibration case database construction module 1, wherein the historical blasting vibration case database construction module 1 is used to collect blasting vibration case data according to blasting design parameters, geological parameters and blasting particle peak vibration velocity; A blasting vibration case database acquisition module 2, wherein the blasting vibration case database acquisition module 2 is used to perform data preprocessing on the historical blasting vibration case database to obtain a blasting vibration case database; A prediction model input variable determination module 3, the prediction model input variable determination module 3 is used to perform a correlation analysis on the blasting design parameters, geological parameters and the peak vibration velocity of the blasting mass point to determine the prediction model input variables; A prediction model setting module 4, wherein the prediction model setting module 4 integrates three base regressors, namely, a single gradient enhancement regressor, a single random forest regressor and a single decision tree regressor, using a voting algorithm to determine a prediction model; The hyperparameter optimization obtaining module 5 is used to set the hyperparameter optimization target of the prediction model to maximize R 2 , optimizing the hyperparameters of the prediction model by combining genetic algorithm and nested 5-fold cross validation to obtain optimized hyperparameters; A GA-VR model acquisition module 6, wherein the GA-VR model acquisition module 6 is used to train the prediction model based on the optimization hyperparameter and the blasting vibration case database to obtain a trained GA-VR model; A prediction sample data acquisition module 7, wherein the prediction sample data acquisition module 7 is used to obtain prediction sample data according to the prediction model input variables; The prediction value acquisition module 8 is used to input the prediction sample data into the GA-VR model to obtain the prediction value, wherein the prediction value is , It is the predicted value of the peak vibration velocity of the explosion particle output by the GA-VR model.

[0031] Furthermore, the blasting vibration case database acquisition module 2 is used to execute the following method: Obtaining multidimensional preprocessing indicators, wherein the multidimensional preprocessing indicators include duplicate data removal, missing data removal, and abnormal data removal; The historical blasting vibration case database is cleaned according to the multi-dimensional preprocessing index to generate the blasting vibration case database.

[0032] Furthermore, the prediction model input variable determination module 3 is used to perform the following method: Calculate the Pearson correlation coefficient between the peak vibration velocity of the blasting mass point and each blasting design parameter and geological parameter; Counting the attribute distribution of blasting design parameters and geological parameters in the blasting vibration case database; The prediction model input variables are selected according to the Pearson correlation coefficient and the attribute distribution of the blasting design parameters and geological parameters.

[0033] Furthermore, the optimization hyperparameter obtaining module 5 is used to execute the following method: The nested 5-fold cross validation includes an inner loop and an outer loop. The inner loop searches for optimal hyperparameters based on genetic algorithm optimization and provides optimal hyperparameters to the outer loop. The outer loop provides data samples for training to the inner loop and reserves some data samples as a test set to test the effect of the inner loop model.

[0034] Furthermore, the GA-VR model acquisition module 6 is used to perform the following method: When the training is not completed, the hyperparameters of the prediction model are selected, crossed, and mutated according to the genetic algorithm and the nested 5-fold cross validation to obtain the hyperparameters of the updated prediction model; Iteratively optimizing the hyperparameters of the prediction model to obtain iteratively optimized hyperparameters; Based on the iterative optimization of the hyperparameters, training is continued according to the hyperparameters of the updated prediction model.

[0035] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The above describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0036] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

[0037] This specification and the accompanying drawings are merely exemplary illustrations of the present invention and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the scope of the present invention. Thus, the present invention is intended to include such changes and modifications if they fall within the scope of the present invention and its equivalents.

Claims

1. A blasting vibration prediction method based on an integrated algorithm, characterized in that: The method comprises: S1. Collect blasting vibration case data based on blasting design parameters, geological parameters and blasting particle peak vibration velocity, and build a historical blasting vibration case database; S2. Preprocessing the historical blasting vibration case database to obtain a blasting vibration case database; S3. Perform correlation analysis on the blasting design parameters, geological parameters and peak vibration velocity of the blasting particles to determine the input variables of the prediction model; S4. Using the blasting vibration case database to construct an integrated prediction model based on a voting algorithm, wherein the base regressor consists of a single gradient boosting regressor, a single random forest regressor, and a single decision tree regressor; S5. Set the prediction hyperparameter optimization target of the prediction model to maximize R 2 , optimizing the hyperparameters of the prediction model by combining genetic algorithm and nested 5-fold cross validation to obtain optimized hyperparameters; S6. Based on the optimization hyperparameters, the prediction model is trained according to the blasting vibration case database to obtain a trained GA-VR model; S7. According to the prediction model input variables, obtain prediction sample data; S8. Input the predicted sample data into the GA-VR model to obtain the predicted value , It is the peak vibration velocity of the explosion particle output by the GA-VR model.

2. The blasting vibration prediction method based on integrated algorithm according to claim 1 is characterized in that: S2. Preprocessing the historical blasting vibration case database, including: S21. Obtaining multidimensional preprocessing indicators, wherein the multidimensional preprocessing indicators include duplicate data removal, missing data removal, and abnormal data removal; S22. Performing data cleaning on the historical blasting vibration case database according to the multi-dimensional preprocessing index to generate the blasting vibration case database.

3. The blasting vibration prediction method based on integrated algorithm according to claim 1 is characterized in that: S3. Performing correlation analysis on the blasting design parameters, geological parameters and the peak vibration velocity of the blasting mass point to determine the input variables of the prediction model, including: S31. Calculate the Pearson correlation coefficient between the peak vibration velocity of the blasting mass point and each blasting design parameter and geological parameter; S32. Statistical distribution of blasting design parameters and geological parameter data in the blasting vibration case database; S33. Select the prediction model input variables according to the Pearson correlation coefficient and the attribute distribution of the blasting design parameters and geological parameter data.

4. The blasting vibration prediction method based on integrated algorithm according to claim 1 is characterized in that: S4. Using the blasting vibration case database, a prediction model based on an integrated voting algorithm is constructed, comprising: S41. The base regressors trained on the same new data set are integrated using a voting algorithm to obtain an integrated model; the base regressors include a single gradient boosting regressor, a single random forest regressor, and a single decision tree regressor; S42. Divide the input data set, and divide the prediction model input variables into training sets T train and the test set T test , the corresponding outputs are recorded as P train and P test ; S43. Train the base regressor using the dataset T train and P train The base regressors are trained, and the base regressors include a single gradient boosting regressor, a single random forest regressor, and a single decision tree regressor. The trained base regressors are used to predict the test set T test , whose predicted output is recorded as P t1 , P t2 and P t3 ; S44. The key hyperparameters in the voting ensemble model include: the number of decision trees constructed in a single gradient boosting regressor n_estimators, the maximum depth of the decision tree max_depth, the learning rate learning_rate, and the minimum number of samples required for node splitting min_samples_split; the maximum depth of the decision tree max_depth, the number of decision trees constructed n_estimators, and the minimum number of samples required for node splitting min_samples_split in a single random forest regressor; the minimum number of samples required for node splitting min_samples_split in a single decision tree regressor, the maximum number of leaf nodes in the decision tree max_leaf_nodes, the maximum depth of the decision tree max_depth, and the proportion of the three base regressors of the voting algorithm.

5. The blasting vibration prediction method based on integrated algorithm according to claim 1 is characterized in that: S6. Training the prediction model according to the blasting vibration case database includes: S61. When the training is not completed, the hyperparameters of the prediction model are selected, crossed, and mutated according to the genetic algorithm and the nested 5-fold cross validation to obtain the hyperparameters of the updated prediction model; S62. Iteratively optimize the hyperparameters of the prediction model to obtain iteratively optimized hyperparameters; S63. Based on the iterative optimization of hyperparameters, continue training according to the hyperparameters of the updated prediction model.

6. A blasting vibration prediction system based on an integrated algorithm, characterized in that: The system is used to execute the method according to any one of claims 1 to 5, and the system comprises: A historical blasting vibration case database construction module, wherein the historical blasting vibration case database construction module is used to collect blasting vibration case data according to blasting design parameters, geological parameters and blasting particle peak vibration velocity; A blasting vibration case database acquisition module, wherein the blasting vibration case database acquisition module is used to perform data preprocessing on the historical blasting vibration case database to obtain a blasting vibration case database; A prediction model input variable determination module, the prediction model input variable determination module is used to perform a correlation analysis on the blasting design parameters, geological parameters and the peak vibration velocity of the blasting mass point to determine the prediction model input variables; A prediction model setting module, wherein the prediction model setting module integrates three base regressors, namely, a single gradient boosting regressor, a single random forest regressor, and a single decision tree regressor, using a voting algorithm to determine a prediction model; The hyperparameter optimization module is used to set the hyperparameter optimization target of the prediction model to maximize R 2 , optimizing the hyperparameters of the prediction model by combining genetic algorithm and nested 5-fold cross validation to obtain optimized hyperparameters; A GA-VR model acquisition module, wherein the GA-VR model acquisition module is used to train the prediction model based on the optimization hyperparameter and the blasting vibration case database to obtain a trained GA-VR model; A prediction sample data acquisition module, the prediction sample data acquisition module is used to obtain prediction sample data according to the prediction model input variables; A prediction value acquisition module is used to input the prediction sample data into the GA-VR model to obtain a prediction value, wherein the prediction value is , It is the peak vibration velocity of the explosion particle output by the GA-VR model.

Images