Day-ahead transformer load and voltage forecasts for rooftop-solar feeders
TL;DR Feederpeak tells a discom, a day ahead, which distribution transformers and low-voltage feeders will run hot or push voltage too high as rooftop solar spreads. It also gives a spare-capacity number per transformer for approving new rooftop applications. Discoms pay a yearly licence per circle.
Rooftop solar in India reached 33.41 GW by September 2026[1], and it lands unevenly, so one colony can add dozens of systems on a single transformer in a year. At midday power flows backwards and voltage rises, and at night the same transformer carries the air-conditioning load. Simulation studies find overvoltage is usually the first limit on how much solar a feeder can take[2].
Discoms already feel the transformer stress. About 10% of distribution transformers fail each year, roughly 1.3 million units, with overloading among the causes, while the best utilities stay under 0.5%[3]. When a housing society applies for shared rooftop solar, the discom must check the transformer's spare capacity before approving it[4], often with old load data.
What it does, in regulatory terms: forecasting and decision support for discom planners only. Feederpeak does not control inverters or switch equipment, and the discom keeps every operating decision.
It builds a network model of each feeder from the discom's GIS and asset records, then combines it with smart-meter readings and weather forecasts. Physics handles how power and voltage move along the wires, and machine learning handles what people and panels will do tomorrow. The planner gets a daily list of transformers likely to overload or breach voltage limits, plus a per-transformer number showing how much more rooftop solar it can take.
That second number speeds up approvals and shows the discom where upgrades are needed first. Over a season the planner also sees which transformers keep popping up on the list.
The discom pays, usually from its planning or IT budget. Private discoms have bought AI grid software through partners before - one US firm's demand-response programme in Mumbai covered 55,000 homes and 6,000 large customers[5]. State discoms buy through tenders. My working assumption is ₹25–60 lakh a year per circle, and it needs testing.
Smart meters must be read remotely at least daily under the 2023 consumer-rights amendment[6], so interval data now exists on feeders that had none. Rooftop solar growth and CEA's attention to transformer failures[3] give planners a real reason to use it.
Nobody publishes what discoms spend on distribution-grid forecasting in India. The narrow signal is the number of transformers under stress, with about 1.3 million failures a year[3].
Rooftop capacity of 33.41 GW[1] is adjacent context for where that stress will grow.
One meter-analytics firm sells meter-data analytics that flag transformer overload and hidden solar, with named utility clients in North America and Europe[7]. A US grid-software firm works with a Mumbai private discom on demand response[5]. Large grid-software vendors sell planning tools to utilities.
The gap is a day-ahead, transformer-level forecast built for Indian feeders, where asset records are patchy.
| Difficulty | Hard (58/100) |
| Starting capital | ₹2–3.8 Cr |
| Time to launch | 12–18 months |
Two things make this hard. Discom network records are often incomplete, so building a usable feeder model means field checks and a lot of cleanup. And state utilities buy slowly, through tenders that favour vendors with past references, which is a catch-22 for a new company. The forecasting itself is the easier part.
No licence is needed to sell forecasts, but you're working inside a heavily regulated sector, so the discom's rules shape everything.
Building this outside India? The categories above usually have a local equivalent, so check with your own regulator.
Don't let AI make operating decisions. Feederpeak is decision support only, and every switching or approval call stays with the discom's engineers.
Want to work out the AI layer for Feederpeak? Talk to Yash →
Here's what can go south -
I think this fits a founding team with power-systems engineering and discom sales experience. If you're an AI team new to utilities then it's a poor fit, honestly.
Start with one private discom and prove one thing - that next-day overload flags on 200 transformers are right at least 70% of the time.
Plan for roughly ₹2–3.8 Cr, because cleaning feeder data and waiting out slow utility sales cycles takes time and a skilled team.
No licence to sell forecasts, but you need smart-meter data agreements, DPDP processor terms and tender empanelment for state discoms.
Yes, rooftop solar is straining distribution grids in many countries. Utilities everywhere buy slowly, so check local procurement rules and grid connection standards before you start.
The free Feederpeak validation kit has the line-by-line capital breakdown, the effort scoring, the utility data and procurement checklist, a 14-day validation sprint with pass/fail targets, a 90-day launch plan, 10 interview questions for discom planners, and 5 assumptions to test. Download the kit →
Thinking about building Feederpeak or something close to it? The make-or-break work is getting clean feeder data and a first discom willing to pilot, and I can help you plan the route to both. If you need a hand, you know where to find me. Talk to Yash →
For illustration only. Names are fictional and ideas hypothetical; sourced facts are real but may be out of date. Nothing here is legal, financial or professional advice, and Yash Khatri is not personally liable for any use of it. Read the full disclaimer.