Work

PCA-BP Neural Network-Based Mining Cost Forecasting Model for Underground Metal Mines: A Gold Mine Case

Year: 2026

Type: article

Abstract: To achieve scientific cost forecasting, this study investigates structural changes in mining cost driven by the widespread adoption of mechanized mining, increased mining depths, and significant operational variations. Based on the backpropagation (BP) neural network, this study systematically analyzes the cost-composition characteristics of modern mining operations and applies activity-based costing to achieve refined cost accounting for each mining operation unit. Ten key influencing factors, including working space, stope temperature, stope depth, haulage distance, worker seniority and work efficiency, scraper efficiency, equipment service life, fuel and lubricant consumption rates, are identified by analyzing cost variation patterns. Principal component analysis (PCA) is used to reduce the dimensionality of the ten factors to simplify this model and enhance prediction accuracy. The PCA-BP neural network mining cost forecasting model is built with the principal components extracted as input variables. Actual cost data from an underground metal mine in Shandong Province is used for our model training and validation, with adopting linear regression, eXtreme Gradient Boosting (XGBoost), and a traditional BP neural network as the comparison models for performance evaluation. Our prediction results indicate that the PCA-BP model achieves an average relative error of 3.80% and a root mean square error of 1.43, both significantly outperforming the comparison models. The results demonstrate superior predictive accuracy and stability of our model. Validated with data from a typical deep mechanized gold mine in eastern China, the PCA-BP cost forecasting model requires parameter retraining based on local production conditions for applications in other regions. This study confirms that the model aligns well with the cost characteristics of modern underground metal mines and produces effective predictions, offering reliable quantitative support for the development of cost control strategies and optimization of cost planning in mining enterprises.

Source: Applied Sciences

DOI: 10.3390/app16126094