A data-mining approach to predict influent quality

Area of Science:

• Environmental Engineering
• Data Science

Background:

• Accurate prediction of influent water quality is crucial for efficient energy management in wastewater treatment plants.
• Key water quality metrics include carbonaceous biochemical oxygen demand (CBOD), potential of hydrogen (pH), and total suspended solids (TSS).
• Gaps in time-series data due to industrial data acquisition limitations hinder accurate CBOD monitoring.

Purpose of the Study:

• To develop a data-driven approach for time-ahead prediction of CBOD in wastewater treatment.
• To address challenges posed by missing CBOD data in industrial time-series datasets.
• To investigate the impact of seasonality on CBOD prediction models.

Main Methods:

• Utilized four data-mining algorithms: multilayered perceptron (MLP), classification and regression tree (CART), multivariate adaptive regression spline (MARS), and random forest (RF).
• Employed experimental approaches to approximate functional relationships and fill missing CBOD data points.
• Developed prediction models with a maximum prediction horizon of 5 days, incorporating seasonality effects.

Main Results:

• Successfully constructed data-driven models for time-ahead CBOD prediction.
• Demonstrated the feasibility of filling data gaps in CBOD time-series data.
• Evaluated the performance of MLP, CART, MARS, and RF algorithms for this prediction task.

Conclusions:

• The proposed data-driven approach effectively predicts CBOD levels in wastewater influent.
• The methods developed can overcome limitations of industrial data acquisition systems for water quality monitoring.
• Accurate CBOD forecasting up to 5 days ahead supports optimized energy management in wastewater treatment plants.