GridKind

POWER THAT PLANS FOR PEOPLE. An explainable microgrid decision-support twin that tests outages, searches feasible allocations, and exposes reserve and fairness trade-offs.

Comment

Inspiration

An outage is an energy event, but its consequences are human. A clinic can lose refrigeration, a cooling shelter can exceed capacity, and a battery can look efficient while the wrong block goes dark. GridKind lets a community rehearse those trade-offs before an emergency.

What it does

GridKind is an explainable decision-support twin for a neighborhood with solar generation, a community battery, a care clinic, a cooling shelter, and 72 homes. Users can select baseline, heatwave, or island-mode scenarios; adjust battery reserve and care-site priority; and optimize for equity, carbon, or cost.

The prototype calculates a hard energy budget, enumerates more than 3,000 candidate allocations, rejects plans that violate the reserve constraint, and displays the winning clinic, shelter, and household allocations. Its audit trace exposes supply, reserve, candidate count, objective weights, curtailed demand, and the selected plan.

How we built it

The interface is built with React, TypeScript, Vinext, and Vite and deployed on Cloudflare-compatible edge infrastructure. Current Melbourne temperature and cloud cover come from Open-Meteo, with a labelled offline fallback. The optimizer is deterministic and constraint-based so its decisions remain reproducible.

Sensor Lab emulates the small JSON solar payload expected from a future ESP32. It is explicitly labelled simulated; no physical device is claimed.

Challenges

The main challenge was making resilience more than a single efficiency score. We separated critical and flexible loads, made fairness priorities adjustable, and kept the result inspectable instead of hiding it behind an AI label.

Accomplishments

  • Working interactive 24-hour decision-support prototype
  • Live public weather input with transparent fallback
  • Constraint search across thousands of feasible allocations
  • Equity, carbon, and cost objective modes
  • Four-step audit trace and visible curtailed demand
  • Hardware-shaped sensor interface without claiming physical hardware

What we learned

Optimization is more useful when people can inspect the constraint and understand why one plan won. A simulated hardware boundary can also validate integration design before equipment is purchased.

What's next

  1. Connect a physical ESP32 solar monitor.
  2. Add forecast uncertainty and outage-duration probabilities.
  3. Let communities define their own critical sites and fairness rules.
  4. Validate assumptions with energy engineers and emergency planners.

Accuracy note

GridKind is decision-support software, not a live electrical controller. Weather is live when Open-Meteo responds; load, battery, and infrastructure values are documented scenario assumptions.

Built With

Share this project:

Updates