Comparison of effluent estimation methods for municipal and industrial WRRFs

The nutrient removal processes in water resource recovery facilities (WRRF) have complex interactions and often, the scarce instrumentation that suffers from bio-fouling in the harsh conditions. Hence, a simple algorithm would be useful to monitor the status of the instrumentation, and to provide an early warning if maintenance is necessary. This study demonstrates how different scaling methods can enable use of simple linear estimation methods for effluent monitoring in biological treatment processes at municipal and industrial WRRFs. In this work, we develop methodology that encompasses data pre-processing, variable selection, data scaling and estimation algorithms. We compare three scaling methods and three linear modeling methods. The scaling methods consist of normalization, moving average with different window sizes, and nonlinear scaling. The linear estimation methods are partial least squares regression (PLSR), auto-regressive moving average with exogenous inputs (ARMAX) and state-space model (SS). These methods are applied to two case studies at municipal and industrial WRRFs for estimating the nutrient composition of effluent wastewater. The results show that the nonlinear scaling combined with PLSR and SS models provides sufficient dynamic fit to the real data. Hence, the methodology can be further used for development of a monitoring and early warning system.