424 PREDICTIVE MACHINE LEARNING MODEL FOR STRATEGIC HEAP LEACH PAD PLANNING IN OPEN PIT MINES *A. Paravecino1, N. Okada2, Y. Ohtomo2, Y. Kawamura2 1Division of Sustainable Resources Engineering, Hokkaido University, Japan, (*Presenting author: angelo.paravecino@pucp.edu.pe) 2Division of Sustainable Resources Engineering, Hokkaido University, Japan ABSTRACT This study develops a predictive modeling framework for strategic heap leach pad planning in open pit mining operations. An integrated dataset was constructed from a geometallurgical block model, incorporating ore grade attribution, solubility parameters, spatial characteristics, and operational placement records. A linear regression model was implemented using engineered features representing ore placement dynamics, heap geometry, and inventory proxies. The model was evaluated using time-ordered splits to preserve temporal causality. The baseline achieved R² values of 0.64 in training and approximately 0.38 in validation and testing periods, demonstrating stable generalization performance without evidence of severe overfitting. The results indicate that linear models capture a significant portion of production variability; however, residual unexplained variance suggests the need for dynamic and nonlinear modeling approaches. This work establishes a transparent benchmark for future physics-informed and machine learning models aimed at optimizing heap discharge planning and maximizing metal recovery efficiency. KEYWORDS Heap leaching; Machine learning; Production forecasting; Spatial modeling 1. INTRODUCTION Heap leaching is a widely used metallurgical process for the extraction of gold from low-grade ores. In this system, ore is stacked in lifts over an impermeable geomembrane and irrigated with a cyanide solution, allowing dissolved gold to be recovered over time. However, gold production in heap leach operations does not depend solely on the quantity of ore stacked during a given time. Instead, production results from a delayed and cumulative response mechanism governed by: Ore stacking rates, Ore grade and cyanide solubility, Column height and hydraulic flow path, Active inventory within the heap, and the temporal evolution of leaching kinetics. Strategic mine planning requires reliable production forecasting to guide lift sequencing, irrigation allocation, discharge prioritization, and cash flow projections. However, conventional forecasting methods often assume proportionality between recent stacking and production, neglecting internal inventory and spatial variability. This study addresses this limitation by proposing a dynamic, data-driven framework grounded in geometallurgical modeling and machine learning.
RkJQdWJsaXNoZXIy MTM0Mzk2