From Rule-Based to ML-Based Energy Management
Conventional energy management systems operate on Rule-Based Logic — users define every condition manually: switch off air conditioning at 23:00, turn off lights after 30 minutes without motion detection. This approach works reasonably for fixed routines but lacks the flexibility to accommodate the natural variation of real household life.
Next-generation Smart Longevity systems using Machine Learning take a fundamentally different approach. Rather than requiring users to write rules, the system observes, analyses, and identifies usage patterns from actual data, then automatically constructs Automation Logic aligned with each household’s real behaviour. This is the distinction that makes a home genuinely "learn" rather than simply "follow instructions."
The 4–6 Week Learning Timeline
In the initial installation period, the system enters Observation Mode for 4–6 weeks, collecting comprehensive energy data without making any adjustments. Data captured includes the on/off timing for each device, energy consumption by circuit for every time slot, occupancy patterns in each zone, and external variables such as outdoor temperature, humidity, and daylight duration.
After four weeks, a Clustering algorithm groups recorded patterns into distinct Daily Profiles — work-from-home days, days working outside, weekends, and special occasions. The system then begins generating Automation Suggestions that the homeowner can approve or decline. These responses become additional training data, enabling the system to refine its model progressively through weeks five and six — and continuously thereafter.
Anomaly Detection and Vampire Load Identification
One of the highest-value capabilities of ML-Based Energy Systems is Anomaly Detection — identifying consumption patterns that deviate significantly from the established baseline. When an air conditioner draws 40% more energy than its normal profile, the system flags a likely maintenance issue: a dirty filter, low refrigerant, or a failing compressor. Early detection prevents the gradual energy creep that goes unnoticed on monthly bills until it is severe.
Vampire Loads — energy drawn by devices in standby mode, chargers left plugged in, or appliances not fully powered down — typically account for 5–10% of a household’s total electricity consumption. The ML system identifies Vampire Loads by appliance, and can configure Smart Plugs to cut power automatically during confirmed non-use periods, eliminating this waste without any ongoing manual intervention.
Seasonal Adaptation for Thailand’s Climate
Thailand’s three seasons create distinct shifts in household energy behaviour. The hot season from March to May with temperatures of 38–40°C drives maximum air conditioning load. The rainy season from June to October brings high humidity but lower temperatures. The cool season from November to February can see overnight temperatures of 15–20°C in Bangkok, substantially changing air conditioning usage.
A well-designed ML system recognises these Seasonal Patterns and adjusts Baseline Energy Profiles accordingly, preventing false anomaly alerts when consumption rises legitimately with seasonal temperature changes. The system can also generate advance projections — predicting how much higher next month’s electricity bill is likely to be based on current seasonal trajectory, enabling proactive budget management.
Recommendation Engine for Behaviour Change
Beyond automating device operation, the ML system functions as a Recommendation Engine that suggests behaviour adjustments to residents through LINE OA. Rather than abstract percentages, recommendations are framed as specific monetary amounts: raising the bedroom air conditioning setpoint from 22°C to 25°C saves ฿180 per month; shifting laundry to after 22:00 saves an additional ฿90.
Presenting energy data in terms of concrete, real-money savings rather than kilowatt-hours or efficiency ratios creates far stronger motivation for behaviour change. This approach transforms Smart Longevity’s energy goals from an abstract system metric into a tangible, daily-life outcome that residents actively engage with.