TY - JOUR
T1 - Data-mining methods predict chlorine residuals in premise plumbing using low-cost sensors
AU - Saetta, Daniella
AU - Richard, Rain
AU - Leyva, Carlos
AU - Westerhoff, Paul
AU - Boyer, Treavor H.
N1 - Publisher Copyright:
© 2021 American Water Works Association
PY - 2021/1/1
Y1 - 2021/1/1
N2 - Variable water quality within buildings is of increasing concern due to public health impacts (e.g., lead, Legionella pneumophila, Naegleria fowleri, disinfection byproducts). Advances in data acquisition and analytics provide the opportunity to monitor real-time building-wide water quality variability. Accordingly, the goal of this research was to create a water quality sensor platform including data acquisition, storage, and mining methods able to monitor, and ultimately improve, water quality within buildings. The platform was used to monitor water temperature, pH, conductivity, oxidation–reduction potential, dissolved oxygen, and chlorine using sensors only. Other building data infrastructure, specifically Wi-Fi logins by occupants, were used to approximate activity rates and associated water use. An advanced machine-learning technique, gradient boosting machines, predicted the chlorine residuals throughout the building plumbing network better than multivariate linear regression models. Finally, the implications of water quality monitoring on costs, scalability, reliability, human dimensions, regulatory compliance, and future green building designs are considered.
AB - Variable water quality within buildings is of increasing concern due to public health impacts (e.g., lead, Legionella pneumophila, Naegleria fowleri, disinfection byproducts). Advances in data acquisition and analytics provide the opportunity to monitor real-time building-wide water quality variability. Accordingly, the goal of this research was to create a water quality sensor platform including data acquisition, storage, and mining methods able to monitor, and ultimately improve, water quality within buildings. The platform was used to monitor water temperature, pH, conductivity, oxidation–reduction potential, dissolved oxygen, and chlorine using sensors only. Other building data infrastructure, specifically Wi-Fi logins by occupants, were used to approximate activity rates and associated water use. An advanced machine-learning technique, gradient boosting machines, predicted the chlorine residuals throughout the building plumbing network better than multivariate linear regression models. Finally, the implications of water quality monitoring on costs, scalability, reliability, human dimensions, regulatory compliance, and future green building designs are considered.
KW - Water quality
KW - chlorine
KW - data
KW - machine learning
KW - premise plumbing
KW - sensors
UR - http://www.scopus.com/inward/record.url?scp=85107634748&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85107634748&partnerID=8YFLogxK
U2 - 10.1002/aws2.1214
DO - 10.1002/aws2.1214
M3 - Article
AN - SCOPUS:85107634748
SN - 2577-8161
VL - 3
JO - AWWA Water Science
JF - AWWA Water Science
IS - 1
M1 - e1214
ER -