Secondary Use of Domestic Metering Data, a Potential Way of Detecting Anomalies in Households and Inherent Behaviors

Introduction: In smart cities, the volume of data is increasing due to more frequent measurements and a
growing number of metering points. This data can be utilized for various purposes within public infrastructure,
extending beyond the primary reasons for installing connected metering devices. There are different objectives
for using this data, and numerous applications can be identified.[1] In this discussion, we will focus on using
data to detect anomalies as an alarm system, with the goal of ensuring the safety of elderly residents.
Material and Methods: This study investigates the currently available meters while evaluating their data
quality and exploring the potential of existing hardware. It also addresses the latency in data transmission and
its associated limitations. Various architectural solutions are examined and described, focusing on how
existing meters can be utilized to identify anomalies in consumption patterns. To gain practical insights into
the systems, installations, and available data, we concentrated on the Åland Islands, a small community. Two
electricity metering companies provide a limited range of measurement devices, data transfer, and storage
technologies. In terms of water metering, the study specifically examines digital, remotely communicated
meters, focusing on two actively used models and one data storage system [2,3].
Results: The findings suggest that there is a potential secondary application of data for detecting anomalies.
One challenge to address is the latency involved in the transmission of data from the meter to storage, which
must be accessible in a structured data format suitable for processing with Machine Learning algorithms.
Additionally, there is a challenge regarding the precision of measurement timestamps. Electricity metering is
highly precise for reporting usage in defined hourly intervals, while water measurement lacks similar precision
in both its primary data and the timing of measurements. The main issue with inaccurate or missing timestamps
is that the algorithms primarily rely on time series analysis, where accurate timestamps are essential for
effectively detecting anomalies. [2,3,6]
Discussion: This study demonstrates that the architecture of a system designed for anomaly detection is
beneficial across various approaches, particularly for detecting inherent anomalies and their potential
correlation with health data or system states. To address the issues outlined in the results section, further
research on implemented edge machine learning architectures is essential to evaluate the effectiveness of the
proposed design. The underlying concept of utilizing edge machine learning is to execute data processing
close to the measurement device, thus reducing latency and ensuring that all recorded data is timestamped for
the purpose of conducting time series analysis.