Application of Artificial Intelligence (AI) to predict mine water quality, a case study in South Africa
Sakala, Emmanuel; Novhe, Obed; Vadapalli, Viswanath Ravi Kumar
Council for Geoscience, South Africa
This paper presents work done to develop a predictive system in response to the coal mine water pollution in South Africa. Mine-water pollution in particular acid mine drainage (AMD) is one of the major environmental challenges in South Africa and prediction of mine water quality into the future is one of the limitations to design effective intervention options. AMD prediction is conventionally carried out using geochemical modelling tools. However, geochemical models are usually based on certain assumptions and these approximations may overlook some factors affecting the processes in water and demands a lot of inputs and model parameters that are often unknown.
Based on studies done, the parameters (rainfall, air temperature, depth to water table and discharge pH) were selected as key controlling factors and sulphate as sensible indicator of AMD pollution. Historic values of these values at a site are used to train artificial neural network (ANN) system. The long-short-term memory (LSTMs) are used to predict future values of each on the controlling factors parameters, results which are feed into the trained ANN. The output from the ANN are predicted future values of sulphate concentration.
Witkrans discharge in Carolina town of Mpumalanga province of South Africa with water chemistry data collected for the past four years was used for the development of the prediction system. The results shows a slight decrease in sulphate concentration over time. Considering the success of the prediction system at Witkrans discharge site, the system can be scaled up and used for regional or catchment management of water resources. The benefits derived from predictive modelling using LSTM and ANN is critical in designing sustainable pollution treatment solutions.