Classification of heterotrophic plate counts (HPC) in a water distribution network: a fuzzy rule-based approach

Abstract

Heterotrophic plate count (HPC) is one of the most common indicators used to monitor microbiological water quality in distribution networks. This paper applies and compares two fuzzy rule-based models to estimate HPC levels in a distribution network using a limited number of water quality parameters. The proposed fuzzy rule-based models include Mamdani and TSK (Takagi, Sugeno, and Kang) algorithms. The models are discussed through a case study of a distribution network (DN) in Quebec City (Canada). Both models properly estimate when HPC levels (datum per datum) are low, however their predictive ability is limited when HPC levels are high. When the outputs (HPC levels) are converted into four classes and the models are used as “classifiers”, their performances are very good. The average percent deviation is lower for the TSK model (15%) than for Mamdani model (17%). Implemented as “classifiers”, both models can be used for identifying vulnerable locations for microbiological contamination within the DN. Given the complexity of the growth of HPC bacteria in water distribution networks and the involvement of numerous contributory factors; results obtained are “promising”. Nevertheless, strategies to improve the models are also discussed.