A NEURO-PREDICTIVE MODEL OF AN INDUSTRIAL WASTEWATER TREATMENT PROCESS USING PRINCIPAL COMPONENTS ANALYSIS

Abstract

This paper presents a way of predicting the biochemical oxygen demand (BOD) of the output stream of the activated sludge of a food processing industry. A combination of principal components analysis (PCA) and artificial neural networks (ANN) was used to develop the network model structure that contained eight neurons in the hidden layer. PCA was used to preprocess the training data sets comprising of four input variable that have been transformed into four principal components before being fed into a back propagated neural network. A more satisfactory result was obtained from PCA-ANN model with correlation index 0.998 and performance error (MSE) of 0.007 compared with that of ANN model with correlation index 0.82 and performance error (MSE) of 0.056 for the training process. This result shows that preprocessing data will bring about improvement in prediction.