Assessment of a Soft Sensor Approach for Determining Influent Conditions at the MWRDGC Calumet WRP

Many older water reclamation plants (WRPs) are implementing model-based process control to satisfy increasingly stringent effluent requirements while they lower energy costs, but these methods require reliable process information. Soft sensors can help to provide such information by building on easily acquirable and historical data. In this investigation of a soft-sensor approach at a conventional WRP, many historical data were missing, which suggested that a large fraction of the information would be lost if the missing data were not well managed. This study applied an iterated stepwise multiple linear regression (ISMLR) approach to minimize the loss of data and to predict real-time influent ammonia and ${\mathrm{CBOD}}_5$ (5-day carbonaceous biochemical oxygen demand), and future influent flow at the Metropolitan Water Reclamation District of Greater Chicago (MWRDGC) Calumet WRP. Relative to a simple deletion method (which retained about 45% of the daily data), the ISMLR approach successfully retained substantially more information (65–82% of the days) and provided good predictions on real-time ammonia concentrations ($r\approx 0.92$) and future influent flow ($\approx 0.91$). Two additional tests, uncertainty analysis and random missing data, were also conducted, which demonstrated the high reliability and flexibility of the ISMLR approach.