GridMind

The UK’s Clean Power 2030 initiative aims to reach at least 95% clean electricity and 23-27 GW of battery storage by 2030, a substantial increase from the current 4.5 GW. To support this goal, batteries must charge during peak wind generation and discharge during shortages. This requires making vital operational decisions in advance before exact market prices are officially known. To solve this, we developed GridMind. This system predicts British electricity prices for the upcoming 22 hours and determines the optimal half-hourly actions for a 50 MW / 100 MWh battery, providing clear natural language explanations for its choices. Our method utilizes a year of Elexon market data, rigorously cleaned to avoid utilizing future information. We train LightGBM models that are updated bi-weekly. A linear-programming optimizer formulates a feasible strategy based on these predictions, intelligently adjusting it every thirty minutes. All data is securely stored in Tiger Data. We successfully overcame major challenges, including implicit look-ahead biases and market price surges of £50/MWh. Over 78 unobserved days, GridMind generated £541k—nearly five times the return of a static plan—proving that maximizing profitability relies heavily on excellent decision-making rather than solely having the best forecast.