Using Artificial Intelligence to Characterize the Effects of Environmental, Operational and Infrastructure Factors on Customer Complaints in a Western Canadian Distribution System

The relationship between water quality problems and customer complaints are critical to understanding factors that affect customer satisfaction. This is because customer complaints are a useful indicator of water quality degradation in a distribution system (DS) as they provide near real-time information on changes in the aesthetics of the drinking water. While providing safe drinking water is the most important component of water supply, this water is susceptible to deterioration once it leaves the plant and enters the distribution system. Deterioration during conveyance occurs when metals accumulate onto and are released from the internal surface of drinking water pipes through corrosion or through biofilm accumulation and release. Water degradation and customer complaints have found to be clustered together in space and time. By using artificial intelligence, specifically self-organizing maps (SOMs) and adopting a temporal and spatial analysis within defined spatial areas, we can identify potential “hot spots” for changes in water quality in a DWDS. Understanding the role of environmental, operational, and infrastructure factors and how they influence the compounded effect of water degradation factors and customer complaints, can contribute to the improvement in water management, infrastructure services and aid in crisis communications.
The aim of this paper is to use artificial intelligence to predict potential hotspots for water degradation based on customer complaints in a Western Canadian Distribution System. Here, SOMs will be used to identify the factors that drive the degradation of drinking water aesthetics in a distribution system. Historical customer complaints data from 2014 to 2019 will be used to detect relationships between water quality and customer complaints and will use principal component analysis (PCA) to compare clustering results.