David Domínguez Barbero has researched Deep Reinforcement Learning (DRL) techniques for the optimisation of microgrid operation. These techniques have several advantages over current applications where they achieve a much more efficient operation strategy without requiring a very complex infrastructure or other highly expensive and not very scalable systems, making them an interesting alternative due to their applicability and sustainability.

This work has resulted in one published article, another in the second round of review and a third in draft form very close to submission, all in JCR scientific journals. A doctoral thesis and, potentially, other articles and research projects that will grow out of the doctoral thesis:

Optimising a Microgrid System by Deep Reinforcement Learning Techniques (Energies 2020, 13 (11), 2830-1,19)

This paper analyses the application in microgrids of novel Machine Learning techniques in the context of optimisation through Deep Learning algorithms combined with Reinforcement Learning. This algorithm optimises the operation of an isolated microgrid for an infinite horizon by constructing an operation policy. In addition, this work analyses the sensitivity of the size of the information buffer stored at each time instant.

Twin-Delayed Deep Deterministic Policy Gradient Algorithm for the Energy Management of Microgrids (under review)

A continuation of previous work. In this case, the policy has access to the entire continuous range of decision variables, as opposed to the previous case where only a discrete set of possible values were candidates for microgrid operation. In particular, the diesel cluster could only take three different values in its operation.

Impact of a Non-Linear Battery Model in a Microgrid using Deep Reinforcement Learning (in preparation)

In the same line of research, DRL techniques can find optimal solutions in non-convex spaces, outperforming classical optimisation techniques that lack this property. In this case, several components of the microgrid exhibit non-linear behaviour, making the application of these DRL techniques more attractive. In particular, these advantages are studied when considering the nonlinear dynamics of lithium-ion batteries.