GridMind

GridMind is an open-source tool that computes cost-optimal EV charging schedules with convex optimization and emits them as OCPP-native SetChargingProfile messages. It sits in the gap between OCPP libraries, which parse the protocol but do not optimize, and energy optimization tools, which optimize but know nothing about OCPP.

The problem

OCPP defines smart charging, but the spec stops at the message format. Deciding when to charge, given time-of-use prices, a shared feeder limit, and each vehicle's deadline, is an optimization problem the protocol leaves to you. Most deployments fall back to charging the moment a car plugs in, which is the worst case for both cost and grid load.

Approach

GridMind models charging as a linear program in CVXPY. The objective minimises energy cost against a price signal; the constraints encode per-connector power limits, battery capacity and target state of charge, session arrival and departure windows, and a shared feeder ceiling across a fleet. The solver returns a per-interval power schedule, which is then serialised into OCPP 1.6 or 2.0.1 SetChargingProfile payloads ready to send to a charge point.

What it does

• Single-EV optimization against time-of-use or flat pricing
• Fleet scheduling under a shared feeder capacity constraint
• Demand response: shifting or curtailing load in response to grid signals
• Strategy comparison of optimized versus uncontrolled charging, with cost and peak-load metrics
• Simulation over synthetic sessions, plus a CLI for configuration and validation

Why it matters

It connects the two halves of my work: the OCPP integration experience from onboarding thirty-plus chargers, and the optimization mindset from ML research. This is the layer that turns a protocol that can do smart charging into a system that actually schedules it well.

View the source on GitHub.