Predicting Effluent Quality Parameters Using Ensemble Models, Artificial Neural Networks and Naked Mole-Rat Algorithm

Abstract

A reliable simulation of wastewater effluent parameters is important for reducing the operational costs of a wastewater treatment plant. In this study, optimized artificial neural networks and inclusive multiple models are used to predict the effluent biochemical oxygen demand, chemical oxygen demand and total suspended solids of a WWTPs in Shahrekord basin, Iran. The influent quality parameters (CODinf, BODinf, TSSinf and PHinf) are used as inputs to the models. The naked mole-rat algorithm is used to tune ANN parameters. This investigation compares the capabilities of ANN-NMRA with those of ANN-firefly, ANN-sine cosine algorithm, ANN-genetic algorithm and ANN models. The output of hybrid and standalone models is incorporated into an ANN model as an IMM model. Several individual models are used in an IMM model for predicting outputs. Hence, an IMM model increases the accuracy of individual models. In this study, ANN models are modified by using a goodness factor to reduce computational time. This paper presents a new preprocessing method for selecting the best input combination and analyzes the uncertainty of model and input parameters. IMM, ANN-NMRA, ANN-SCA, ANN-FFA, ANN-GA and ANN models achieve MAEs of 0.789, 0.998, 1.19, 1.26, 1.34 and 1.40 mg/L for predicting BODeff in the testing stage, respectively. The IMM model has the highest accuracy for predicting CODeff and TSSeff. A good factor reduces the computational time of ANN models by removing redundant hidden neurons. The uncertainty analysis results show that model parameters provide higher uncertainty than input parameters.

Keywords

• effluent quality parameters
• artificial neural network models
• uncertainty
• optimization algorithms
• wastewater treatment plant