Abstract

In the domain of water treatment, improving the efficiency of wastewater treatment plants (WWTPs) has highlighted the need to model certain concentrations and variables characteristic of effluents. Artificial intelligence is an effective tool for monitoring WWTPs and modeling their complex processes. This chapter presents an application of multilayer perceptron (MLP) neural network MLP with two learning algorithms [conjugate gradient (CG) and Broyden–Fletcher–Goldfarb–Shanno (BFGS)] and support vector machine (SVM) model for predicting efficiency of WWTP in terms of effluent chemical oxygen demand (CODeff) and total suspended solids (SSeff). The dataset includes a total of 295 data points that are divided into two phases: training and testing. On the basis of inputs combination, different models are trained and the good fit input combination is examined. The performances of different developed models during the two phases (train and test) are evaluated from a comparison between the observed CODeff, and SSeff values and the predicted CODeff and SSeff values to define the good fit prediction model. Two performance indices like root mean square error (RMSE) and coefficient of determination (R2) are estimated for all developed models (SVM, MLPCG, and MLPBFGS). In the testing phase, the SVM model has an RMSE for SSeff and CODeff varying from 1.375 to 2.384 mg L−1. Using MLPCG model, the RMSE is reduced from 1.154 to 2.038 mg L−1. The RMSE is more reduced with MLPBFGS model (from 0.941 to 1.982 mg L−1). The results showed that MLPBFGS model may be a very useful tool for perfect prediction and process control of WWTPs.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call