An Artificial Neural Network Framework for Prediction of Chemical Oxygen Demand in Wastewater Treatment Plant

Abstract

Determining treated effluent parameters in WWTPs is crucial for assessing operational efficiency, ensuring compliance with environmental regulations, and promoting sustainable and responsible wastewater treatment practices. Conventionally, standard laboratory analysis techniques are used to characterize the wastewater. Artificial Neural Network analysis is a promising tool to predict the wastewater characteristics, as a substitute to tedious laboratory techniques. They enable proactive decision-making and contributes to the overall effectiveness of the treatment processes. This paper constructs various regression models based on neural networks within the domain of traditional machine learning. The main objective of this paper is to design neural network-based regression models to predict Chemical Oxygen Demand (COD), a key parameter characterizing treated effluent of a domestic wastewater treatment plant. Date used in the study is the augmented data of influent and effluent characteristics values from a wastewater treatment plant. The strength of neural networks lies in their ability to capture complex, non-linear relationships in data and make them suitable for predicting wastewater parameters that may have intricate dependencies on input variables. In this paper, the KMeans cluster with Principal Component Analysis (PCA) is used for feature selection and anomaly detection to increase the performance of the regression model. Various algorithms used include Multilayer Perceptron Network, Polynomial Regression with a Neural Network, Generalized Additive Models (GAM) with a Neural Network, Radial Basis Function Network and Artificial Neural Networks (ANNs) with Genetic Algorithms (GAs) with different Hyperparameter methods for prediction. Based on the results, it can be concluded that Average of all five neural network models predicted COD value with Mean Square Error of 0.0015 and R-Square of 0.8522.