Predicting and Forecasting Mine Water Parameters Using a Hybrid Intelligent System

Water treatment plants need to stock chemicals and have enough energy as well as human resources to operate reliably. To avoid a process interruption, proper planning of these resources is imperative. Therefore, a scientifically based, practical tool to predict and forecast relevant water parameters will help plant operators to know in advance which chemicals and methods are necessary for polluted water management and treatment. This study aims to develop a system to predict and forecast mine water parameters using electrical conductivity (EC) and pH of mining influenced water from the Acid Mine Drainage treatment plant in Springs, South Africa as an example. Three machine learning algorithms, namely random forest regression, gradient boosting regression and artificial neural network (ANN) were compared to find the best learning model to be used for predictive analysis. These models were developed using historical data of the years 2016 to 2021. Input variables of the models are turbidity, total dissolved solids, SO4 and Fe, with EC and pH being the target outputs. Results of the models have been compared with the measured data on the basis of the mean absolute error and root mean square error. The results show that random forest and gradient boosting models perform better than the ANN model, and thus these models were deployed as a web application. The Long Short-Term Memory technique was used to forecast the input parameter values for 60 days, and these values were used to get the future values for EC and pH for the same period.