Application of Machine Learning for Operational Forecasting and Compliance Control in Underground Mining

Valentin W.; Valverde C.; Bautista I. · Proceedings of the International Conference on Soft Computing and Machine Intelligence, ISCMI (2025) · DOI: 10.1109/iscmi67495.2025.11358442

Resumen

In underground mining operations, achieving daily physical targets - such as meters advanced or tons transported - is essential to ensure process efficiency and continuity. Many operations still use spreadsheets and manual workflows, delaying deviation detection and hindering decisions. This article presents the design and preliminary validation of an intelligent operational control system for underground mining, integrating hierarchical KPI visualization, automated alerts, and forecasting through advanced machine learning techniques. The system was developed as a Minimum Viable Product (MVP) for the "Americana"unit of the Peruvian mining company Alpayana, using a modular service-oriented architecture (SOA + C4) that consolidates data from CIR (Dispatch) and SAP into dashboards tailored to user roles. The forecasting module applies a Stacking Ensemble model with Random Forest, LightGBM, and LSSVM as base learners, and XGBoost as the meta-learner - chosen for their robustness, accuracy, and adaptability to highly variable operational contexts. Initial validation showed reduced workload, better traceability, and faster deviation detection. Its hybrid deployment (IIS for frontend/backend, and ML + ETL microservices in Docker) ensures scalability, resilience, and alignment with the corporate IT ecosystem. Currently under development, the system is scheduled for deployment in June 2025, with quantitative evaluation and progressive institutionalization planned for subsequent quarters.

Instituciones: Universidad Peruana de Ciencias Aplicadas