In open-pit mining, the selection of deep-hole bench blasting parameters is a critical process that directly influences the uniformity of blasting fragmentation distribution and the morphology of the muckpile. This selection has a cascading effect on the operational efficiency, cost management and safety of subsequent processes such as loading transportation and crushing. Consequently accurately determining appropriate blasting parameters under complex geological conditions to achieve an optimal balance between blasting effectiveness and economic benefits has been a longstanding challenge for researchers and engineering professionals in the field. To address this fundamental issue, this paper introduces a method for predicting and optimizing deep-hole bench blasting parameters in open-pit mines. Utilizing an improved Dung Beetle Optimizer （NDBO） algorithm -enhanced Extreme Learning Machine （ELM）, this approach offers a novel technical pathway for determining blasting parameters. The original DBO algorithm is enhanced by incorporating a dynamic inertia weight adjustment mechanism. Which balances the algorithm’s global exploration capability in the initial stages with its local exploitation capability in the later stages. In order to address convergence stagnation within the algorithm, a chaotic perturbation strategy is employed to perturb the population, thereby facilitating the algorithm’s escape from local optima. This approach markedly enhances the solution accuracy and stability of the algorithm， culminating in the development of the improved NDBO algorithm. During the model construction phase, the NDBO algorithm is integrated with the Extreme Learning Machine （ELM） to optimize the core parameters of the ELM model, leveraging the global optimization capabilities of the NDBO algorithm. Specifically, the NDBO algorithm optimizes the input layer weights. hidden layer thresholds, and output layer weights of the ELM enabling adaptive adjustment of model parameters and the construction of the NDBO-ELM blasting parameter prediction model. To evaluate the superiority of the proposed model, comparative experiments are conducted against traditional intelligent optimization models, including PSO-LSSVM （Particle Swarm Optimization-Least Squares Support Vector Machine） and PSO-ELM （Particle Swarm Optimization-Extreme Learning Machine）. A quantitative analysis was conducted utilizing four key performance indicators: Root Mean Square Error （RMSE）, Mean Absolute Error（MAE）, Mean Absolute Percentage Error （MAPE） and the Coefficient of Determination （R²）. The experimental findings indicate that the NDBO-ELM model demonstrates substantial advantages in predicting the three primary blasting parameters. Specifically, for hole spacing prediction, the MAE, RMSE, MAPE and R² values are 0.14, 0.18, 0.02 and 0.91, respectively. For row spacing prediction, these metrics are 0.14, 0.16, 0.03 and 0.94, and for specific charge prediction. they are 0.006, 0.007, 0.01 and 0.93. The low levels of error indicators and the coefficients of determination, which are consistently close to or exceed 0.9, suggest that the model possesses exceptionally high predictive accuracy and stability. It effectively captures the intricate nonlinear relationships between blasting parameters, geological conditions, and blasting outcomes. Building on the verified efficacy of the NDBO-ELM model, this study further develops an optimization system for deep-hole bench blasting parameters in open-pit mining operations. Anchored in the fundamental design principles of “data-driven, intelligent optimization, and closed-loop iteration” the system incorporates two essential functional modules: blasting data collection and intelligent parameter optimization. It establishes a comprehensive closed-loop management framework encompassing “data collection-model training-parameter recommendation-effect feedback.”

http://www.goldsci.ac.cn/CN/10.11872/j.issn.1005-2518.2026.03.348