Prediction of wastewater treatment plant performance using multilinear regression and artificial neural networks

In this study, modeling of Konya wastewater treatment plant was studied by using multilinear regression and artificial neural network with different architectures in SPSS and MATLAB software. All data were obtained from wastewater treatment plant of Konya during daily records over four month. Treatment efficiency of the plant was determined by taking into account the input values of pH, temperature, COD, TSS and BOD with output values of COD. To compare the performance of the model, coefficient of determination (R²) and Mean Squared Error (MSE) were used. In Multilinear regression method, to understand the effects of the tested parameters, regression function was developed. The highest prediction efficiencies was obtained two hidden layers in Artificial Neural Network models. According to the modeling study, Artificial Neural Network models responded more satisfactory results than Multilinear Regression model.

I. Introduction

During the recent decades, environmental researches related to wastewater treatment plant have become popular due to rising concern. The developments in modeling methods make them possible to use complex system. Because of the complex system, the performance of digesters under changing organic loading rates is unpredictable [1]. Several successful models have been published for waste water treatment plants (WWTP) [2]–[5]. These models are about forecasting WWTP effluent parameters properties, process control of WWTP, predictions of WWTP parameters. Some of the parameters used for the model validation were temperature, total suspended solids (TSS), pH, chemical oxygen demand (COD), biogas production, Biological Oxygen Demand (BOD) and volatile solids. Unsuitable operation of a WWTP may cause to happen disquieting problem such as environmental and public health. Its effluent to a receiving water body can cause a broad spectrum disease to human beings. To control of a WWTP for predict the WWTP performance, the robust models can be developed using past observation data. [6].