Existing machine learning and data-driven models are normally established upon big data without knowing physical properties of a given system (e.g. water/wastewater treatment unit). To conquer this problem of “Big data, Small Task”, this study explored an innovative approach to obtain high spatiotemporal-resolution data and equip models with physical meanings, through which prediction accuracy of data-driven models could be effectively enhanced as “Small Data, Big Task”. Specifically, a novel physics-informed data-driven model was developed based on datasets obtained from flat milli-electrode array (MEA) sensors (conductivity, pH and temperature) in a continuously-stirred tank reactor (CSTR) to predict system performance. Three of each type of MEA sensors were deployed along with the CSTR depth to real-time in situ profile heterogeneity (Fig.1a). Shock solution was individually injected into the CSTR using a syringe.  For model development, the reactor was divided into four compartments (top, middle, low and mixing zone) (Fig.1b). Reduced-order multi-compartment models were established based on the transport mechanism of Na+ (conductivity), H+ (pH) and heat (Temperature). Rigorous deterministic global dynamic optimization algorithms were deployed to validate these data-driven models by solving the corresponding parameter estimation problems. This unique physics-informed data-driven modeling effectively determined the best-possible predictive capability for mixing dynamics under shocks (Fig.1b), represented by excellent fitting to MEA sensor data without sacrificing physical meanings (Fig.1c). In contrast, machine learning Eureqa® solely depended on data suffered from lack of physical meanings (Fig.1d) and could not be used for system prediction or control. By integrating two-fold physical-data properties, this powerful modeling could predict physicochemical state of water/wastewater systems, especially under transient shocks, and will ultimately enhance system resilience and enable energy-saving operation.

Novel physics-informed data-driven modeling to predict physiochemical state of water/wastewater system

Macromolecular Chemistry: The Second Century

[Author Benefits](https://scimeetings.acs.org/?utm_source=pubs_content_marketing&utm_medium=website&utm_campaign=0320_MCF_NPI_Launch_Spring_Homepage&ref=pubs_content_marketing)

[How to use SciMeetings](https://storage.googleapis.com/dl.morressier.com/acs/SciMeetings-ACSSpring2020-HowToGuide_M3020-Final.pdf)

[Frequently Asked Questions](https://storage.googleapis.com/dl.morressier.com/acs/SciMtng_ACSNMSp2020_FAQ_M3120_Final.pdf)

ACS Spring 2020 National Meeting & Expo

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.