Exploring wastewater treatment plant operation and performance using k-means and Gaussian mixture clustering.

Efficient wastewater treatment is critical to protecting the environment and minimizing human impacts on natural watercourses. Optimizing wastewater treatment to meet compliance objectives while avoiding excessive energy use can be challenging, particularly for plants relying on exclusively manual operation. Even when large volumes of data are available, it can be challenging to derive useful insight from this data to better understand plant behaviours. This study presents the use of unsupervised clustering as an approach to improve operators' and engineers' understanding of WWTPs, demonstrating how clusters can identify operational 'fingerprints'. These fingerprints enable a clearer understanding of different processes by reducing numerous different signals into a small set of easily interpretable operating states. We use these fingerprints to identify upstream and downstream associations with WWTP operation and performance and identify optimal operating strategies for secondary treatment to minimize energy use while meeting effluent compliance objectives. We also explore how these fingerprints can be used to identify when plants may be at risk of critical events like exceedances or bypasses. Finally, we include a comparison of two different clustering algorithms (k-means clustering and Gaussian mixtures clustering) and explore the advantages and disadvantages of both algorithms for understanding WWTP operation performance.