Graduation Year

2024

Document Type

Thesis

Degree

M.S.E.V.

Degree Name

MS in Environmental Engr. (M.S.E.V.)

Degree Granting Department

Civil and Environmental Engineering

Major Professor

Qiong Zhang, Ph.D.

Committee Member

Katherine Alfredo, Ph.D.

Committee Member

Hui Wang, Ph.D.

Committee Member

Hung Quoc Nguyen, Ph.D.

Keywords

Machine Learning, Water Treatment, Chemical Consumption, Energy Use, Sustainability

Abstract

As part of the UN Sustainable Development Goals (SDG6), access to clean water for the entire population and sustainable management of water resources have been stressed. Demand for water is increasing due to population growth, economic development, water quality degradation, and climate change. More water needs to be produced for the community at lower treatment costs. The chemical dosage and energy intensity are two important factors that influence water treatment costs and sustainable management of resources as well.

Previously, chemical dosage modeling to reduce treatment costs have been explored but there have been more focus in coagulant dosage modeling. Other chemicals used in the treatment plants were taken less into consideration. Moreover, there have been less work on developing energy intensity models. This research uses Linear Regression, Random Forest (RF), XGBoost, Deep Learning and Time Series Analysis to develop data-driven models for predicting chemical dosage and energy intensity for different supply sources of Tampa Bay Water. Such models can be used to minimize the treatment costs by allocating different sources for meeting the demand.

The result from this study indicates that production largely influences the energy intensity of different treatment plants with a correlation coefficient of 0.93 for desalination, 0.98 for groundwater and 0.27 for surface water. For energy intensity, Random Forest model provided an R2 value of 85%, XGBoost gave the highest R2 of 52% for surface water and Linear Regression gave an R2 of 82% for groundwater. Furthermore, some water quality parameters influenced the dosage of certain chemicals based on correlation analysis and the order of treatment processes. Ferric Sulfate, the coagulant used for surface water treatment had a strong correlation with color where the model gave an R2 of 67% from Random Forest. An R2 of 61% was obtained for ammonium sulfate dosage in groundwater where another chemical, sodium hypochlorite was used as input in XGBoost.

Overall, this research provides different data-driven models for predicting chemical dosage and energy intensity of various sources of water supply. Models that gave high accuracy could be used for decision-making, such as reallocating water resources accordingly to reduce treatment cost while at the same time meeting water demand.

Download

Share

COinS