The Innovate UK-funded phase of Oversight, Syntax Genie's AI-powered mini-grid management platform. Forecasts of demand and solar generation drive a control engine that schedules the battery and grid. Designed for rural communities, it was piloted on a hybrid mini-grid at the Alta Vision office building in Colombo: 20 kW solar, 2 kW wind, 50 kWh battery storage and 10 kW diesel backup.

• ★Machine-learning model design and the project's research component.
• ★The Azure platform: IoT Hub ingestion, data, model training and hosting.
• ★Worked with stakeholders and the smart-plug hardware team, as part of the project team.
• ★Outputs: an arXiv survey and a poster at All-Energy 2024, Glasgow (below).

The Alta Vision Solar–RGU partnership that began with this project won the Inward Investment Impact award at the Scottish Knowledge Exchange Awards 2026.

Built a new dataset from a working mini-grid in Colombo — 1,002,195 records, 239 variables, 449 days at five-minute resolution, joined with NASA POWER weather data — and used it to compare forecasting models for energy management.

• ★Compared Prophet, NeuralProphet, LSTM with auto-encoder, and hybrid CNN-LSTM with auto-encoder on a chronological split, with paired t-tests and Diebold–Mariano tests.
• ★The simpler LSTM with auto-encoder beat the hybrid model on unseen data (MAE 176–204 W vs 232–266 W; R² 0.97 for both) at a fraction of the training time.
• ★The models drive an energy management system piloted with a Siemens S7-200 PLC.

A survey of how AI enhances standalone and grid-connected mini-grids to deliver reliable energy access in remote communities. Reviews forecasting approaches — from statistical methods to machine learning and hybrid models — for predicting renewable generation and demand, surveys practical tools like Prophet and N-BEATS, and identifies open research gaps for real-world deployment.

Poster presenting the AI-powered mini-grid management system developed in Sri Lanka — forecasting supply and demand to optimise renewables, battery storage and diesel backup.

Short-term demand and generation forecasting for a working hybrid mini-grid in Colombo, comparing statistical and deep-learning models with formal significance tests, and using the best model to drive a PLC-piloted energy management system.