AeroClim.
Set your location & weather baseline — optional; the offline demo is ready
Search and select your city, then load weather. Without internet, set the manual values instead.
Weather requests send the entered city or coordinates to Open-Meteo. The demo needs no account, API key or ESP connection.Demo data: synthetic local readings and modeled cooling, with regional weather or manual inputs.
See heat before it arrives
The error guide is not a confidence interval. Model skill is evaluated only on generated data.
Four surfaces. Different conditions.
Plan cooling around the forecast
Planned cooling intensity
| Six-hour modeled outcome | Reactive | Optimized |
|---|
Both schedules use the same forecast and simplified room model. Cost and comfort can trade off; savings are not guaranteed.
Why this plan?
Template-based explanation of calculated results; not an LLM response.Model evidence, assumptions & demo notes
Calculated test errors — synthetic data only
| Horizon | Ridge MAE | Persistence MAE | Regional MAE | Test rows |
|---|
Three ridge regressions are trained in this browser on generated temperature-offset history. Inputs include present offset, trend, hour, surface properties, humidity and future regional temperature. There are no prewritten accuracy or savings results.
Training and testing are separated by time and forecast horizon. Both ridge and regional baselines receive the generated future regional temperature. Real weather forecast error, hardware error and transfer to real sites are not validated here. These numbers must not be presented as field accuracy.
The anomaly detector uses Holt level/trend residuals and requires two consecutive high residuals. The planner uses a beam search through off, half-power and full-power actions. Only the forecasting component learns regression coefficients.
Transparent assumptions
- Four synthetic nodes; no USB, MQTT, ESP or sensor connection.
- Room thermal time constant: 4 hours. Extra heat gain: 0.15°C/h. Full cooling effect: 2.6°C/h. Full electrical draw: 2.2 kW.
- Illustrative tariffs: ₹8/kWh, rising to ₹14/kWh from scenario hour +2 to +4. These are not your actual local tariffs.
- Reactive thermostat switches on at the comfort ceiling and off 1°C below it. Optimizer trades off cost, overheating, overcooling and switching.
- Humidity and light are illustrative; no gas, particulate, CO₂ or carbon-emissions claims are generated.
- Location selects regional weather, not a measurement of your room. Manual values are assumptions.
Weather and geocoding: Open-Meteo (CC BY 4.0). Online loading is optional; API availability is not guaranteed. Export run saves inputs, assumptions, generated results, metrics and model coefficients.
1. Prepare. Set your location or manual baseline. Keep the simulated-data label visible. Open Fullscreen and rehearse once.
2. Record. Start your screen recorder, then the 90-second tour. Read or paraphrase the narration below the dashboard. Show your assembled PCB separately as hardware awaiting integration.
3. Explain. Present this as a software simulation prototype. Show calculated synthetic evaluation; describe real sensor validation as the next milestone.