Multimodal Utility Data for Appliance Recognition: A Case Study with Rule-Based Algorithms

Area of Science:

• Energy systems analysis
• Smart home technology
• Data analytics

Background:

• Appliance recognition from aggregate data is difficult due to real-world conditions like noise and concurrent device usage.
• Existing studies often use idealized datasets, not reflecting actual deployment challenges.
• Multimodal utility data (electricity, water, gas) collected at building entry points offers a realistic data source.

Purpose of the Study:

• To investigate appliance recognition using real multimodal utility data under realistic deployment conditions.
• To evaluate the effectiveness of transparent, rule-based detectors for appliance recognition.
• To identify the potential and limitations of rule-based detection in imperfect sensing environments.

Main Methods:

• Collected real multimodal utility data (electricity, water, gas) over six weeks.
• Developed and evaluated transparent, rule-based detectors for four household appliances.
• Focused on exploiting temporal dependencies between modalities and robustness to sensing imperfections.

Main Results:

• Achieved reliable detection for water-related appliances: washing machines (22/30 cycles), dishwashers (19/21 cycles), and tumble dryers (23/27 cycles).
• Highlighted limitations in detecting short, high-power events like kettle usage.
• Demonstrated the potential and constraints of rule-based detection in realistic, imperfect sensing scenarios.

Conclusions:

• Rule-based detection is a viable approach for specific appliance types under real-world conditions.
• Further research is needed to improve detection of short, high-power events.
• The study provides a baseline for hybrid systems combining rules with data-driven methods for enhanced appliance recognition.