Electronic Thesis and Dissertation Repository

Thesis Format

Integrated Article

Degree

Master of Engineering Science

Program

Civil and Environmental Engineering

Supervisor

Dagnew, Martha

Abstract

The aeration process is the largest energy consumer in wastewater treatment plants (WWTPs), and the optimization of the process based on computational models can offer significant savings for the plant. Recent theoretical developments have revealed that many of the parameters commonly assumed as constants in aeration modelling, in fact, have a dynamic nature; however, there still lacks a universal way to model these factors in an easy, accurate and timely manner. This work proposed a machine learning-based modelling approach to offer real-time estimations of the oxygen transfer rate, airflow demand, and energy consumption.

Utilizing the field data collected from Adelaide WWTP (London, Ontario, Canada), the study developed and screened a combination of modelling approaches and input parameters for optimum predictive power under different data availability conditions.

The results demonstrated that the machine learning models provided significantly higher predictive power than the traditional mechanism-based models. These models can provide informative predictions of the aeration parameters with only operational parameters and limited knowledge about the underlying mechanisms of the system. When integrated with the theoretical equations, the models still produce reasonable estimations without losing interpretability. The present finding confirms using the machine learning modelling approach on dynamic factors involved in the aeration process to be feasible and effective. It calls for further investigation into such methods to explore more in the field of wastewater modelling.

Summary for Lay Audience

Just like all dogs go to doggy heaven, all drains in our city lead to a wastewater treatment plant, where the microorganisms break down the pollutants and turn the wastewater back into clean water for us to use. The aeration process is the part in which we pump air into the water to let the microbes breathe and do their jobs. The more air they get, the happier they are and often the cleaner the resulting water can be. However, pumping air is a costly process, so our plants want to minimize the input air without harming the outflow water quality. One potential approach is to adjust the airflow pumps based on need. With the water flow rate and amount of pollutants changing, it is difficult to determine the need manually, but this can be quickly done if we have a mathematical model for estimations. In this study, we present a new way to make accurate and timely estimates of the many factors involved in the aeration process we need to achieve this goal.

While most previous researchers focused on describing mathematically how aeration works, an alternative could be to utilize machine learning methods for predictions. Machine learning models train themselves automatically through experience; however, we cannot easily explain the underlying reasoning of their predictions. To make up for this, we combined the two approaches and tested their performance under different scenarios.

We found that the machine learning models doubtlessly work better most of the time, but the combined models are also generally acceptable and easier for human to understand. We found some interesting relationships between parameters that we never thought of, and we are hoping to have them studied further by other inspired researchers.

To sum up, our study offered a new way to increase the accuracy of our various predictions in the aeration process, to have better control over the energy use in the pumping systems, and hopefully to save a fortune in treating our waste soon.

Download

Share

COinS