A hybrid modeling, monitoring and control approach for wastewater treatment plants

点击翻译按钮获取中文摘要

Wastewater treatment plants are currently designed using fairly detailed first principles models that are able to capture the dominant dynamics of the process, but not necessarily predict the variables very accurately. The calibration of these models for a new plant site remains a significant challenge. The lack of precise models means that reliable monitoring and optimum operation of these plants is not readily achieved. Plant operation data is becoming increasingly more available, and efforts have been made to use completely data driven approaches to model the plant behaviour. The lack of first principles information in the data driven approaches, however, makes the data driven models susceptible to overfitting and not amenable to extrapolation. Motivated by these considerations, the present proposal will develop hybrid models that seamlessly synergies data driven approaches with first principles models in a way to leverage the information present in the first principles model. This project is a followup to a successful engage grant where a proof of concept for the proposed idea was demonstrated. The present proposal will build on that with real case studies, and generalize to significantly expanded applications of monitoring and feedback control. The direct benefit to the industrial partner Hydromantis is an immensely broadened clientele. Given the broad nature of the tool being developed, additional benefit to Canadian economy will be realized through adaptation of the approach to a broad ranges of production facilities ranging from petrochemicals to pharmaceuticals to steel manufacturing.