This project proposes a Hybrid Predictive Framework for indoor climate management using minimal IoT infrastructure. While traditional smart home systems rely on expensive sensor grids, this study investigates the "Predictive Ceiling" of a single DHT22 sensor node.
Our research identifies the transition point where simple Persistence Logic (Reactive) hands over control to SARIMA-based Modeling (Proactive). This framework is designed for "Cold-Start" scenarios, requiring only 7 days of historical data to begin optimizing energy consumption.
The system utilizes a Hybrid Control Loop that combines statistical forecasting (SARIMA) with immediate sensor-triggered overrides for real-time responsiveness.
To provide transparency into the AI's decision-making, the hardware utilizes a three-tier signaling system:
- 🔴 Red LED (Steady) | IDLE STATE: The environment is stable. The appliance (Aircon/Fan) is OFF to maximize energy conservation.
- 🟡 Yellow LED (Blinking) | PREDICTIVE AWARENESS: The SARIMA model has detected a rising thermal trend within the 30-60 minute horizon. The system is preparing to engage.
- 🟢 Green LED (Steady) | ACTIVE COOLING: The appliance is ON. This is triggered either Proactively (to mitigate heat before it peaks) or Reactively (if a sudden heat spike/heater is detected).
We utilize an Ablation Study to compare three logic layers across 10, 30, and 60-minute horizons:
- Persistence (Naive): The baseline "yesterday = today" reactive logic.
- SARIMA (Univariate): Captures thermal inertia and periodic room cycles.
- SARIMAX (Multivariate): Evaluating Humidity as a lead indicator vs. stochastic noise.
- Hardware: Single DHT22 (Temp/Humidity) via ESP32.
- Sampling: 5-minute intervals (300s).
- Dataset: 2,017 observations (Jan 23–30, 2026).
- Validation: Performance is benchmarked against the sensor’s physical resolution (0.1°C) and accuracy (±0.5°C).
-
Persistence Supremacy: In stable indoor micro-climates, the current state is an elite 10-minute predictor (
$MAE \approx 0.02°C$ ). - The Proactive Edge: While SARIMA has a higher MAE than Persistence, it provides Lead Time. It identifies a "Heat Slope" up to 60 minutes before it happens, allowing for low-intensity pre-cooling.
- Humidity Paradox: In residential settings, adding humidity data (ARIMAX) introduced noise rather than precision, suggesting univariate models are more efficient for edge deployment.
By shifting from reactive "Full-Power" cooling to proactive "Low-Intensity" pre-cooling, this framework reduces peak load demand on local grids using sub-$10 hardware.
Enables smart-home retrofitting for older residential structures without requiring invasive multi-sensor installations or months of data collection.