Estimation of wastewater process parameters using neural networks

Abstract

The use of on-line process analyzers for continuous measurement of wastewater components is often accompanied by unpredictable breakdown or necessary maintenance work at the analyzers, leading to fault time in the measurements. In this paper a new method is presented, allowing the approximate calculation or estimation of those process parameters which are temporarily not available. The calculation is based on auxiliary parameters; the results can be used for process control. Therefore, besides the output of the process analyzers, easily and directly measurable auxiliary parameters are determined. The correlations between these auxiliary parameters and the process parameters actually of interest (COD, NH4, etc.) are detected and used for the estimation of the process parameters in case of a breakdown. Information processing is executed by a neural network, enabling the detection of non-linear static or dynamic correlations, based only on information given by the measured values. The network is trained by data recorded before the breakdown of an on-line analyzer. By this, an optimal adaptation to the current wastewater composition is possible. This method was tested on a municipal wastewater treatment plant near Siegen (Germany). The results obtained are presented in this paper.