Applying support vector machines for predicting the impact of raw effluents variation in a wastewater treatment plant

In various field, companies use data mining techniques to assist them in decision-making processes. Among the various application fields we can find the biology and environment domains, in particular approaching several issues related to wastewater treatment plants. Treatment plants are characterized by having several treatment stages for removing solids, organic matter and nutrients, among other things. All this involves very dynamic and complex process that must be handled efficiently to ensure an effluent with good quality. The prediction of the treated wastewater quality, based on the measured inflow parameters, allows for the evaluation of the performance of the treatment and yet to obtain useful information for a better control of the entire WWTP infrastructure. In this paper we explored some data mining techniques for prediction, namely the ones that use regression models, in order to predict concentrations of some quality parameters, like the Biochemical Oxygen Demand or the Total Suspended Solids. The regression techniques used herein were based on support vector machines, more particularly support vector regression and in one of its variants: sequential minimal optimization