In response to the growing demand for intelligent water resource management, this thesis investigates advanced AI methods to analyze, cluster, and monitor residential water consumption patterns using data from IoT-enabled smart meters. The primary goal is to segment households into interpretable behavioral clusters and track how these groups evolve over time, offering both operational insight and policy relevance to water providers.
This work conducts a comprehensive and comparative study of time-series representation learning and clustering techniques, focusing on how consumption patterns of Consumer groups can be captured, grouped, and interpreted over time. A key contribution is the examination of the temporal stability and interpretability of these clusters, evaluating whether meaningful behavioral types persist across different modeling strategies, and identifying when households transition from one usage pattern to another. Such transition detection can provide early signals of behavioral change and enable timely, targeted responses.
The thesis is structured around the following key work packages:
- Method Development, Benchmarking & Evaluation: Designing, comparing, and evaluating time-series representation and clustering techniques on residential water consumption datasets to ensure robust, explainable consumer segmentation.
- Temporal Tracking & Transition Detection: Developing approaches to monitor household consumption over time and systematically detect and interpret shifts between behavioral clusters.
- Interpretability & Consistency Analysis: Investigating the stability and meaning of identified clusters across different methods and timeframes to ensure actionable insights.
A potential use case of this research is the design of adaptive tariff models and other demand management strategies informed by cluster membership and behavioral change, supporting sustainable and efficient water consumption.