IMPACT OF CLIMATIC VARIABLES ON INFLOW VOLUMES TO WASTEWATER TREATMENT PLANTS USING MACHINE LEARNING TECHNIQUES

Abstract

Although the precipitation has decreased over the last decade, intense rainfall events have become very frequent in India. It is generally assumed that the rainfall events do not affect the sewage inflow volume to the wastewater treatment plants having separate sewer system (Sanitary sewers). This study aims at determining the effect of Climatic Variables mainly Rainfall and Temperature to the wastewater inflow to the wastewater treatment plants with Separate Sewer system using Machine Learning methods. In our attempt to determine the relation between the Climatic Variables and Sewage Inflow volumes to the wastewater treatment plants, K-means clustering, Regression trees, Logistic Regression, Random Forests, Local regression and Artificial neural network were used. Models were applied on the daily wastewater inflow values received by IIT Roorkee 3 MLD Sewage treatment plant, Saliyar, nearby Roorkee, Uttarakhand 33 MLD Sewage treatment Plant and Colorado, US 25 MGD (94.635 MLD) STP, all having Sanitary sewer system and daily Rainfall and Temperature values for Roorkee were collected from National institute of Hydrology, Roorkee and Daily Precipitation and Temperature values for Colorado, US were taken from National Oceanic and Atmospheric Administration (NOAA) website. Regression trees were used to see the effect of climatic variables i.e. Rainfall and Temperature to the inflow to the treatment plants. Median correct characterization by regression trees for IIT Roorkee STP was around 66%, for Saliyar it was around 72% and for Colorado, US STP it was around 86%. Logistic regression and Random forest classification were used to predict the risk of surge of inflow to the plants and mean accuracy for both the models for both IIT Roorkee and Saliyar STP was around 85% and for Colorado, US STP it was around 97%. Local regression, Random forest regression and Artificial neural networks were used to predict the sewage inflow to the plants. Local Regression and Random forest were found to perform better than Artificial neural networks and showed median R-squared value of around 74% for IIT Roorkee STP, 80% for Saliyar STP and around 97% for Colorado, US STP. Considering that all collection systems feeding into a Wastewater treatment plant in India exclusively consist of sanitary sewers, this finding points to substantial storm water ingress into the sanitary sewers as well. The usage of machine learning models to demonstrate the effect of rainfall, predicting the risk associated and predicting the sewage influent values for certain values of Rainfall and Temperature, can be used for better designing of the wastewater treatment plant.