Online machine learning model accurately predicts wastewater inflow flow rates

Accurately predicting the inflow flow rate to a wastewater treatment facility is important for proper operation of the treatment facility. Influent refers to the untreated water that enters the plant. Accurately predicting inlet flow rates allows plant operators to plan efficient use of resources.

In previous studies predicting inflow flows, data-driven models have proven to be effective tools. However, most of these early studies focused on batch learning, which proved inadequate for wastewater forecasting during the COVID-19 era, when inflow patterns have changed significantly.

Batch learning, or offline learning, trains a machine learning model from data in batches, and the data is collected over time. In online or streaming learning, the model is trained as new data arrives. Batch learning models typically run faster and require fewer computational resources, but they tend to be less flexible than online learning models when processing large and changing datasets.

Traditional batch learning prediction models are not suitable for prediction problems where input-output relationships change. During the pandemic, the shortcomings of batch learning approaches became more apparent as input-output relationships changed due to COVID-19-related lockdowns. The team turned to online learning models to determine whether they could overcome some of these limitations.

“We leveraged new machine learning techniques to enhance our ability to predict wastewater inflow flows under the COVID-19 lockdown situation,” said Pengxiao Zhou, a civil engineer at McMaster University. A potential application of this study is that the developed model can be integrated into commercial wastewater modeling software.

The online learning models developed by the team are called Adaptive Random Forest, Adaptive K-Nearest Neighbors, and Adaptive Multilayer Recognition. These are based on traditional batch learning models known as random forests, K-nearest neighbors, and multilayer recognition.

The team used a newly developed online model to predict changing inflow flows due to COVID-19. The pandemic has had a huge impact on people’s daily lives. Lockdowns that closed schools, non-essential services and recreational facilities changed the behavior and movement of billions of people. These changes affected sewage treatment plants.The research will be published in a journal Frontiers of environmental science and engineering.

The team developed the model using three to four years of hourly influent flow and weather data collected from two wastewater treatment plants in Canada. They compared the online learning model they developed with each traditional batch learning model to predict the inflow flow rates to the two plants.

The team used two different scenarios. In one scenario he had a 24 hour forecast and in the other he had no lead time forecast. Their online learning model produced accurate predictions under changing data patterns. They were able to efficiently handle continuous, large-scale, and impactful data streams. The team found that the online learning model outperformed the batch learning model.

“The proposed new online learning model can provide stronger decision-making support to wastewater utilities and managers to cope with changing inflow patterns due to emergencies such as COVID-19.” said Mr. Zhou.

Looking to the future, the team’s future research will include more case studies and consider more predictive scenarios to further validate the developed model. “The ultimate goal is to provide reliable tools for wastewater management and promote the development of wastewater intelligence,” Zhou said.