Detection of Anomalies in Daily Activities Using Data from Smart Meters.

Area of Science:

• Electrical Engineering
• Computer Science
• Health Informatics

Background:

• Smart meters are increasingly deployed for energy management and offer potential for non-intrusive monitoring of household activities.
• Analyzing energy consumption patterns can provide insights into occupant behavior and well-being.

Purpose of the Study:

• To design a short-term behavioral alarm generator using smart meter energy consumption data.
• To evaluate different intelligent systems for predicting hourly energy consumption and detecting activity anomalies.

Main Methods:

• Collected household energy consumption data disaggregated per appliance for six months.
• Trained and compared four intelligent systems (Recurrent Neural Network, Convolutional Neural Network, Random Forest, Decision Tree) for consumption prediction.
• Statistically analyzed predictions against actual consumption to detect anomalies in sleeping, breakfast, and lunch activities.

Main Results:

• The Recurrent Neural Network achieved an F1-score of 0.8 in detecting anomalies in daily activities.
• This approach outperformed other methods in predicting energy consumption and identifying behavioral deviations.
• The system demonstrated the feasibility of using smart meter data for real-time activity monitoring.

Conclusions:

• Smart meter data processing enables the development of effective behavioral alarm systems.
• The proposed method, particularly using Recurrent Neural Networks, shows promise for applications in ambient assisted living.
• This technology can contribute to enhanced health and social care through non-intrusive monitoring.