Online Prediction of Influent Characteristics for Wastewater Treatment Plants Management Using Adaptive Recursive NNARMAX Model

摘要

A novel technique for the nonlinear modeling and online prediction of incoming influent characteristics of an activated sludge wastewater treatment (AS-WWTP) is presented in this paper. The nonlinear modelling and online prediction in the presence of disturbances is achieved using an online adaptive recursive least squares (ARLS) algorithm to the nonlinear model identification formulated in this paper. The performance of the proposed ARLS algorithm is compared with the so-called incremental backpropagation (INCBP) which is also an online identification. These two algorithms are validated by one-step, five-step ahead prediction methods as well as the Akaike's method to estimate the final prediction error (AFPE) of the regularized criterion. Furthermore, the validation results show the superior performance of the proposed ARLS algorithm in terms of much smaller prediction errors when compared to the INCBP algorithm. The results from the incoming influent characteristics predictions show three scenarios, namely: high toxic, low toxic and acceptable toxic levels of the incoming influent. The proposed techniques and algorithms can be adapted and deployed for the modeling and prediction of an incoming influent (sewage) for industrial WWTP management systems.