Neural networks for prediction of wastewater treatment plant influent disturbances

Abstract

In order to develop an effective control strategy for the activated sludge process (ASP) of a wastewater treatment plant, an understanding of the nature of the influent load disturbances to the wastewater treatment plant is necessary. The wastewater treatment processes are dynamic and the interrelationships between variables are very complex. The values of the influent disturbances are usually measured off-line in a laboratory, as there are still no reliable on-line sensors available. This work proposes development of a neural network model for prediction of the values of the influent disturbances, which ultimately affect the activated sludge process. Three different dynamic multilayer perceptron feed-forward neural network models and three recurrent neural networks are developed for the prediction of the influent disturbances of chemical oxygen demand (COD), total Kjeldahl nitrogen (TKN) and flowrate respectively. The predictive performance of the multi-layer perceptron is compared to that of the recurrent neural network.