Determining the best and simple intelligent models for evaluating BOD5 of Ahvaz wastewater treatment plant

Abstract

The biochemical oxygen demand (BOD5) could be used as an indication of wastewater treatment quality, but measuring BOD5 is very time-consuming and costly. Ahvaz wastewater treatment plant (A-WWTP) plays a pivotal role in reducing the input load to the Karun River and it is very important to check its efficiency. Thus, the most critical parameters affecting the BOD5 were determined using the linear regression and stepwise method. The capability of the multivariate linear regression model (MLR), feed-forward artificial neural network (FF-ANN), and adaptive neuro-fuzzy inference system (ANFIS) were investigated with different architectures and inputs to predict the effluent BOD5 of A-WWTP (for daily and monthly modes). These architectures had two, three, four, or five inputs. The results of the MLR revealed that the maximum correlation coefficients (R) for training and testing were 0.916 and 0.864 (daily), and 0.809 and 0.793 on a monthly basis, respectively. The maximum R in FF-ANN for training and testing was 0.960 and 0.906 (daily basis), and 0.921 and 0.849 (monthly basis), respectively. Meanwhile, the maximum R in ANFIS for training and testing was 0.980 and 0.933 daily, and 0.968 and 0.927 monthly, respectively. The results indicated that the three models are appropriate, but the ANFIS is a more accurate model. In addition, based on conditions and available wastewater qualitative parameters, all of the architectures can be used to estimate the output BOD5.