Reinforcement learning for solar battery dispatch
GréineQ replays a historical solar-battery day in 30-minute steps. At each step the agent sees solar generation, household demand, battery charge level, and grid price — then chooses hold, charge from solar, discharge to the home, grid-charge (buy), or export (sell). Grid-charge and export enable price arbitrage: buy low, use or sell later.
Operate the battery for 48 half-hour steps. Lowest adjusted electricity cost wins.
You always see a realistic 4h forecast. Privileged opponents may use a true direction signal — oversight demo.
| Controller | Net cost | Savings vs no battery | Gap to oracle |
|---|