Predictive Model for Water Consumption in a Copper Mineral Concentrator Plant Located in a Desert Area Using Machine Learning

In this study, water consumption predictions are made based on data obtained from two copper mineral concentration plants located in the northern region of Chile, Antofagasta Region, measured during a year of mining operation. The area to which this operation belongs is characterized by a desert climate with average annual rainfall of less than 50 mm and maximum and minimum temperatures of 29 °C and 4 °C for summer, and 26 °C and −5 °C for winter, measured during 2022. To perform the predictions, a database corresponding to two concentration plants with daily measurements for a year was used, which were analyzed using four regression models using Machine Learning (ML) in Python: Support Vector Regressor (SVR), Extreme Gradient Boost (XGBoost), Artificial Neural Network (ANN), and Random Forest Regressor (RF). The predictions obtained by each of the ML models were studied using cross-validation hyperparameter tuning and identifying the variable with the greatest impact. The model with the best prediction results was ANN, as it yielded the lowest relative error in the predictions.