Energy Islands Management in a Decarbonized World. 483-709

Mission:

As renewable energy generation units are being deployed worldwide, distribution grids are facing integration challenges. A promising solution are microgrids (MGs), i.e. energy islands which can be defined as self-managed small power systems consisting of loads, generation units and storage, that may allow intentional islanding and connection to the main grid. The candidate proposal will investigate intelligent cost-efficient management of MGs, where each MG can be operated as a single controllable entity.

Up-to-day, two main classes of control approaches can be identified, rule-based strategies and (predictive) optimization approaches. In the first one, actions are taken based on a (heuristic) predefined response pattern to a set of monitored conditions. In the second one, decisions are the variables of an optimization problem, defined over a given time horizon. Although rule-based approaches have been traditionally adopted due to their simplicity and effectiveness, optimization-based approaches that use different mathematical techniques are gaining increased attention. The latest is further growing thanks to the application of machine learning techniques

Following the machine-learning optimization-based approaches trend, the main objective of the thesis will be the use of machine learning techniques to discover effective control laws and rule-based laws which allows us to perform optimal (in the sense of minimum footprint) operation of MG. The base-line approach will be to extend the idea of generation of models of a system from data to the case of control laws.

Apart from standard MG figures of merit such as quality energy and standard energy balance metrics, the cost function should also account for climate risks, production and consumption forecasts profiles, and production costs and energy markets tendencies. Then, the controller must have a flexible and general representation so that a search algorithm may cover a wider space of possible controllers. A machine learning algorithm will be chosen to discover the most suitable control law in terms of structure and parameters through some training procedure involving data.

Functions:

• The candidate will have to analyse, propose and develop advanced algorithms to support smart actions during each step of the MG life-cycle using AI and machine-learning tools.
• The candidate will be trained in the operation of power systems and MGs. He/She should prove certain knowledge on AI and related topics.
• The candidate should be creative, collaborative, open-minded and passionate about new energy challenges.