Value or Z-Score based detection
- Slope of usage change in summer
- Afternoon Usage to Evening Usage Ratio
- Morning Usage to Evening Usage Ratio
- Appliance Usage
Heuristic Rules
- Find > 6 kWh usage in moderate weather: 70-75F.
- Ans: 07/01/13 at 4 AM.
- Python Time-Series resample at custom-interval (e.g. 5min, 1hr, Mondays) with how=sum allows building powerful queries.
Usage Distribution Model
- Histogram of per-minute usage values shows multi-modal distribution.
- K-Means or GMM clustering techniques are considered.
- But need to vary the #of clusters from 2 to 8 and determine the best-fit #.
- Future work to test G-Means.