Energy systems generate large volumes of heterogeneous information, including meter and sensor measurements, weather forecasts, electricity-market data, network models, operational procedures, technical reports and regulatory documents. Converting this information into timely and reliable decisions normally requires specialised modelling tools and considerable domain expertise.

Large language models (LLMs) and agentic AI create new opportunities to make energy-system analysis and decision-making more accessible, automated and adaptive. Rather than only generating text, an AI agent can interpret a task, retrieve relevant information, analyse data, call forecasting or optimisation tools, interact with simulations and explain its recommendations. Multiple specialised agents may also collaborate to address complex problems involving energy users, renewable generators, storage systems, electricity networks and markets.

However, general-purpose AI models are not inherently reliable for energy applications. They may produce factually incorrect answers, overlook physical and operational constraints, misinterpret numerical data or recommend actions that are unsafe or infeasible. Energy applications therefore require AI systems that are grounded in trusted data and domain knowledge, connected to appropriate analytical tools, and evaluated for accuracy, robustness, security and explainability.

This project investigates how LLMs, foundation models and agentic AI can support the analysis, planning, operation and management of energy systems. Depending on the student’s interests, the project may address applications such as:

• developing an energy-domain assistant that answers technical questions using trusted documents and data;
• using retrieval-augmented generation to interpret market rules, standards, operating procedures or energy policies;
• building AI agents that analyse energy data and automatically select appropriate forecasting or analytical tools;
• integrating LLMs with optimisation, power-flow, simulation or digital-twin tools;
• developing agents for renewable-energy and battery planning, scheduling or operational decision support;
• coordinating distributed energy resources, electric vehicles, microgrids or virtual power plants through multiple AI agents;
• using multimodal AI to reason across text, time-series data, network information, tables, diagrams or images;
• generating and evaluating energy scenarios for planning, resilience or emergency response;
• creating conversational tools that provide personalised and actionable household energy advice;
• extracting structured knowledge from technical reports, market notices and regulatory documents;
• explaining forecasting, optimisation or control outcomes to operators and other stakeholders;
• examining collaboration between human experts and AI agents in high-stakes energy decisions; and
• evaluating hallucination, reliability, privacy, cybersecurity, fairness and safety in energy-focused AI systems.

Each student will select a focused problem and develop a prototype or analytical framework using suitable energy datasets, documents, simulation environments or software tools.

#sustainability