A power outage risk prediction app, built for small industrial businesses in Coimbatore, Tamil Nadu, as a concept for Qutuhal InnovateX 2.0 (Frontier Innovators track).
Ohm takes live weather reports, local outage records, and self-reported infrastructure information and distills it into an easy-to-read Low / Elevated / High indicator.
• All processing is done on the device; no backend is implemented or planned. User location, infrastructure information, and all analytics are local to the device. • Heuristic, not ML: weather, history, and infrastructure factors are combined with hard-coded weights (see
core/riskEngine.ts) rather than a trained model. There is not yet enough data on local patterns to train something responsibly, and the long-term goal is to upgrade to something like XGBoost after enough local history has been gathered. • Generally falls back on cached data if a data source is unavailable, rather than failing catastrophically.
Source Description Notes
Open-Meteo Live weather report Shows hourly wind, precipitation, and storm chance. Free, requires no API key. Actual API.
NammaMap outage feed Scheduled TANGEDCO outages Unofficial; may be down or change formats unexpectedly. There is currently a reported bug (fetch() call from expo start --web) where this fails to load in a browser, but works when loaded directly in a browser tab - likely a CORS issue, and should behave normally on a native mobile build.
On-device confirmation log Past outage frequency for this device Always starts empty for a fresh install. Local to this device only; no backend to aggregate histories across users.
Self-reported infrastructure Feeder type, connection age, etc. Lowest priority factor, as this is the user-submitted data, and may be noisy.
• Currently no official feed for Indian power outages; using an unofficial circular aggregator with no uptime guarantees. Can only predict weather-related or pattern-based outages; not equipment failures, theft, or human error. • Local history is per-device; there is no backend to let the app learn from other users' data, for privacy reasons. This also means that a user's risk history only tracks their own outages, not regional ones they might be affected by. • Only effective in Coimbatore; other regions in Tamil Nadu or India have not been considered. The outage parser and threshold weights have only been tested in Coimbatore. • The weights in riskEngine.ts are not based on any research; the papers cited below only show that such a model could work, not how to implement it.
Allcott et al (NBER / American Economic Review, 2016) - electricity shortages reduce average Indian plant revenue by 5-10%; businesses without backup generators are worst hit. Lee et al (ORNL, IEEE IRI 2023) - machine learning can predict outage risk based on historical data (EAGLE-I database) + National Weather Service alerts. Shows that outage prediction is possible using pattern recognition; does not specify how to implement it. Other references to India's electricity problems, blackout economics, small-business impacts - see the project's tech spec for a more detailed list.
npm install
npx expo start
Then scan the QR code with
Expo Go on a phone, or type w in the terminal to test in a browser (note: reported issues with the outage feed fetcher in browser mode)
Make sure to use a recent Expo SDK version, as some mobile devices may refuse to load the project if your expo go app is too old. If this happens, try installing the sdk-specific builds directly from https://expo.dev/go?sdkVersion=&platform=android or ios , rather than relying on the google play / app store listings, which may not always have the latest sdk.