AI Feeder: planning distribution capex with load flow you can check

AI Feeder is the distribution network planner Solnix built for power utilities. It maps the medium-voltage feeder network, forecasts load growth for every distribution transformer and feeder, analyses sectionalizing, load switching and hosting capacity, and ranks capital works into a phased plan that fits the budget, with a written justification for each item.

AI network mapping rebuilds feeder connectivity from asset coordinates and scores each span's confidence, so low-confidence spans are flagged for survey instead of silently trusted. Consumer-to-transformer indexing uses smart-meter voltages to check which transformer each consumer is really on. The network reads and writes OpenDSS and GeoJSON, and imports transformer meter data as CSV.

Load growth is forecast per transformer with a blend of trend, seasonal and Holt-Winters models, weighted by how well each one back-tests on that transformer's own history. The output is a P50 and P90 peak, growth rate, time to overload and a health index, plus growth hotspots and a feeder-by-year view.

Load flow runs a backward/forward sweep at each feeder's real voltage, from 4.16 to 33 kV, with a voltage profile against the planning limit. It is cross-validated against EPRI OpenDSS to within 0.0016 per unit, and the OpenDSS cross-check runs in continuous integration, so a change that breaks agreement fails the build.

For reliability, AI Feeder computes SAIFI, SAIDI and energy not served by zone and recommends where to place sectionalizers and reclosers. Every tie switch is tested for load transfers and N-1 restoration is reported. Hosting capacity shows how much solar and new load, such as EV charging, each transformer can take and which limit binds first.

Candidate works (transformer additions or augmentation, feeder bifurcation and reconductoring) are each valued and scored. An optimiser selects the set that fits the budget and phases it by year, then writes a justification report for every item. Plans move from draft to submitted to approved, and an approver cannot approve their own plan.

Six AI agents sit in a chat panel in the dashboard: a planning copilot, a capex advisor, a scenario analyst, a data and mapping agent, a plan reviewer and a hosting and reliability agent. They work through role-scoped tools that call the same engines as the screens, so an answer in chat matches the numbers on the page. A public demo with sample data is live.

Frequently asked questions

Found this useful?

Share the key finding on X.