AI applied to energy systems
Energy systems produce more data than almost any other part of an industrial operation, and use less of it. This programme is about closing that gap carefully.
Why this programme exists separately
Most people writing about artificial intelligence do not understand energy systems, and most people writing about energy do not understand machine learning. The result is a literature that is either technically competent about models and naïve about tariffs, metering and grid behaviour, or the reverse.
The intersection is where the practical value sits, and where independent material is hardest to find. That is why it gets its own programme rather than being a subheading of either neighbour.
What we cover
Consumption forecasting
Short-horizon load forecasting is one of the few industrial machine-learning applications with an unambiguous, calculable payoff, provided you can convert a forecast error into money under your actual tariff and balancing arrangements. Much published work reports error metrics without ever making that conversion, which leaves the reader unable to judge whether the improvement matters.
Load and tariff optimisation
Shifting flexible load against a time-varying tariff, managing peak demand charges, and deciding which processes are genuinely shiftable rather than theoretically shiftable. The constraint is almost never the optimiser; it is the production schedule and the people who own it.
Photovoltaic asset performance
Detecting underperformance in an installed array — soiling, shading, string faults, inverter clipping, degradation — from generation data rather than site visits. This is anomaly detection with an unusually clear ground truth, which makes it a good first application for an organisation building capability.
Storage dispatch
When a battery earns its cost back, and under which tariff structures it does not. Dispatch strategy interacts with degradation, warranty terms and self-consumption in ways that simple payback models routinely miss.
What we insist on in this programme
- Every accuracy figure is converted into an operational or financial consequence. A percentage on its own is not a result.
- Data quality is stated before method. An impressive model on unvalidated meter data is a confident error.
- Bulgarian conditions — irradiation, tariff structure, grid arrangements — rather than figures imported from another market.
- Where a spreadsheet would do the job, we say so.
Who this is for
- Energy and facility managers deciding whether monitoring investment will change anything.
- Industrial operations teams with metering data and no clear use for it.
- Municipal energy staff managing building portfolios.
- Analysts and consultants needing methods grounded in local conditions.