Energy Explained
The Role of Artificial Intelligence in Energy Management
Artificial intelligence is doing real work in energy management, but not the work the marketing suggests. Here is the honest division of labour.
By Kettle Energy, Engineering team7 min read
Almost every energy technology company now describes its product as AI-powered. Some of that is genuine, some is a rebranding of statistical methods that have existed for decades, and a little of it is simply marketing. It is worth separating the three.
Where it genuinely helps
Forecasting
Predicting the next 48 hours of wholesale prices, grid carbon intensity, solar output and household demand is a pattern-recognition problem with abundant historical data. This is exactly the shape of problem machine learning is good at, and the accuracy gains over simple rules are real and measurable.
Scheduling under uncertainty
Given imperfect forecasts, deciding when each battery in a fleet should charge is an optimisation problem with competing objectives: cost, carbon, battery health and resident comfort. Learned policies can outperform fixed rules, particularly when conditions are unusual.
Anomaly detection
A battery behaving slightly differently from its peers is often the first sign of a developing fault. Detecting that drift automatically, across hundreds of units, is a task no human operator could do by inspection.
Where it does not help
- Safety limits. Charge rates, temperature thresholds and isolation must be deterministic rules, not learned behaviour.
- Billing and settlement. These need auditable arithmetic, not a model whose reasoning cannot be reproduced.
- Explaining results to residents and landlords. Evaluation should rest on measured half-hourly data that anyone can check.
“A model can decide when to charge. It should never decide whether something is safe.”
The test to apply
When a supplier claims AI capability, ask what decision the model makes, what data it was trained on, and what happens when it is wrong. A credible answer describes a bounded decision with a deterministic fallback. A vague answer usually means the phrase is decorative.
