Comparison of an ANN-based and a regression tree based model to predict effluent pollutants concentrations based on commonly measured influent pollutants concentrations

Abstract

Data-driven models that can predict effluent pollutant concentrations in Sewage Treatment Plants (STP) are increasingly gaining popularity in the research literature. They have emerged as a viable alternative to parameter intensive, rigorous mechanistic models. This prediction methodology helps the plant operators monitor plant operations, take timely corrective action, and manage the effluent concentrations accordingly as per the norms. The data used in this work were obtained from a significant conventional treatment plant in Bahadarabad, near Haridwar. Daily records of Biochemical oxygen demand (BOD), Total suspended solids (TSS), Chemical oxygen demand (COD), pH through various stages of the treatment process over ten months were obtained from the plant laboratory. This analysis initially considers the relevance of ANN techniques to predict effluent concentrations of BOD, TSS and COD. Due to the flexibility of ANN models in capturing the dependencies between a wide variety of input and output data, it is used to predict effluent pollutant concentrations. Due to the categorical nature of the observed effluent values, Regression trees are also used to predict the effluent concentrations of commonly occurring pollutants. A comparison of the performance of both these models and relevant discussions are provided in this report.