Online machine learning for stream wastewater influent flow rate prediction under unprecedented emergencies

Accurately predicting the incoming flow rate is necessary for operators and managers at wastewater treatment plants. In simple terms, influent flow is the untreated water entering a treatment plant. This prediction is closely tied to the characteristics of the wastewater, such as biochemical oxygen demand (BOD), total suspended solids (TSS), and pH. 

Previous research demonstrated that data-driven models effectively predicted influent flow rates. However, many of these studies focused on batch learning, where data is collected over time, and the machine learning model is trained in batches. This approach is good, but this approach needed to be revised, particularly in the COVID-19 era, when influential patterns saw significant changes.

In machine learning, batch learning involves processing data in discrete chunks over time. Conversely, online learning involves continuously training the model as new data becomes available. During COVID-19, the limitations of batch learning approaches became more pronounced due to the changes in input-output relationships brought about by COVID-19 lockdowns. In response, the team redirected their focus towards exploring online learning models to ascertain their potential in overcoming these limitations.

Pengxiao Zhou, a civil engineer at McMaster University, explained that they employed innovative machine learning techniques to enhance the capacity to predict wastewater influent flow rates, especially within the unique context of the COVID-19 lockdown situation.

The researchers compared the performance of conventional batch learning models, including Random Forest, K-Nearest Neighbors, and Multi-Layer Perceptron, with their respective online learning counterparts—Adaptive Random Forest, Adaptive K-Nearest Neighbors, and Adaptive Multi-Layer Perceptron—in predicting influent flow rates at two wastewater treatment plants in Canada.

Online learning models consistently outperformed conventional batch learning models across various scenarios, exhibiting the highest R2 values, the lowest Mean Absolute Percentage Error (MAPE), and the lowest Root Mean Square Error (RMSE). In all instances, the R2 values for the testing dataset, predicting influent flow rates 24 hours in advance, were noteworthy. 

The researchers found that these online learning models effectively provided reliable predictions amid dynamic data patterns. They exhibit efficiency in handling continuous and substantial influent data streams. 

The team crafted their models by leveraging three to four years of hourly influent flow rate data and meteorological data obtained from two wastewater treatment plants in Canada. They conducted a comparative analysis by pitting the online learning models against their corresponding conventional batch learning models to forecast influent flow rates at these two plants.

Pengxiao Zhou remarked that the new online learning models could provide more robust decision support for wastewater operators or managers to cope with changing influent patterns due to emergencies such as COVID-19.

To further validate the efficacy of the built models, the team’s future efforts will involve conducting more case studies and exploring a wider range of prediction scenarios.

Check out the Paper and Reference Article. All Credit For This Research Goes To the Researchers on This Project. Also, don’t forget to join our 32k+ ML SubReddit, 40k+ Facebook Community, Discord Channel, and Email Newsletter, where we share the latest AI research news, cool AI projects, and more.

If you like our work, you will love our newsletter..

We are also on WhatsApp. Join our AI Channel on Whatsapp..