Advances in soft sensors for wastewater treatment plants: A systematic review

Abstract

The on-line monitoring of wastewater treatment plant (WWTP) operations is a challenge due to interference and breakdown from the harsh conditions endured by sensors in wastewater. To lessen the dependence on hardware sensors, mathematical models have been developed for estimating wastewater parameters. These so-called software (soft) sensors have advanced significantly, from mechanistic modelling to the latest machine learning models. The current review aimed to characterize these advancements in WWTP soft sensors by (1) identifying the current status of WWTP soft sensors; (2) analyzing the advancements in soft sensor development methods over time; and (3) evaluating WWTP soft sensors in relation to hardware technology. It is difficult to define an all-encompassing ‘state-of-the-art’ owing to significant variations in the physical and statistical properties of different WWTPs. However, the study was able to evaluate the effectiveness of these methods in specific contexts, based on the statistical properties of the dataset used for soft sensor development. It found that, although neural networks have remained the dominant methodology for soft sensor development since the early 2000s, some decision tree-based approaches have shown promising performance and enhanced robustness. It also highlights the importance of adjunct statistical methods for handling multicollinearity and noise, which are common problems in WWTP datasets. Opportunities to use soft sensor modelling approaches to enhance hardware sensor performance have also been identified. Continuous improvements in the reliability and range of measurement of hardware sensors, are expected to enhance the performance and scope of application of WWTP soft sensors.