Keywords

predictive models; risk; machine learning; drinking water

Start Date

7-7-2022 9:20 AM

End Date

7-7-2022 9:40 AM

Abstract

In the framework of global change, climate change is affecting the quality of surface water. One of its possible effects is the increase of the pathogen load in existing drinking water treatment plants (DWTPs). This topic put a question mark on whether existing utilities are able to handle possible future scenarios, as a result, the need for early warning systems and strategies to detect and respond to acute events in drinking water systems is highlighted. The objective of this study is to develop an early warning system related to the microbiological safety in a DWTP and integrate it as a tool to aid in daily decisionmaking. Among the different machine learning models, the non-parametric approach Bayesian multivariate linear regression (BLR) showed good accuracy on the test data and allowed the quantification of uncertainty in the model’s outputs. The target of the model was Clostridium Perfrigens as an indicator of the presence of Cryptosporidium species in raw water. Several data coming from online sensors are used to predict the expected pathogen load at the DWTP inlet using a machine learning algorithm. Following, the expected pathogen removal along the drinking water treatment process is prescribed from a range of values reported in the literature. Finally, all this information is combined to estimate risk metrics like the disability adjusted life years (DALY). The whole algorithm was applied to the test data and results showed the expected risk of produced water including uncertainty (95% confidence interval) boundaries.

Stream and Session

false

Share

COinS
 
Jul 7th, 9:20 AM Jul 7th, 9:40 AM

Machine learning models as a suitable tool to estimate at real time the microbial risk in drinking water

In the framework of global change, climate change is affecting the quality of surface water. One of its possible effects is the increase of the pathogen load in existing drinking water treatment plants (DWTPs). This topic put a question mark on whether existing utilities are able to handle possible future scenarios, as a result, the need for early warning systems and strategies to detect and respond to acute events in drinking water systems is highlighted. The objective of this study is to develop an early warning system related to the microbiological safety in a DWTP and integrate it as a tool to aid in daily decisionmaking. Among the different machine learning models, the non-parametric approach Bayesian multivariate linear regression (BLR) showed good accuracy on the test data and allowed the quantification of uncertainty in the model’s outputs. The target of the model was Clostridium Perfrigens as an indicator of the presence of Cryptosporidium species in raw water. Several data coming from online sensors are used to predict the expected pathogen load at the DWTP inlet using a machine learning algorithm. Following, the expected pathogen removal along the drinking water treatment process is prescribed from a range of values reported in the literature. Finally, all this information is combined to estimate risk metrics like the disability adjusted life years (DALY). The whole algorithm was applied to the test data and results showed the expected risk of produced water including uncertainty (95% confidence interval) boundaries.