The drive towards sustainability in the energy industry is being addressed by the continued uptake of renewable energy sources and through the reduction of energy consumption by improving efficiency. Introducing intermittent renewable sources, such as wind and solar power, adds more uncertainty to the already complex, multivariate, nonlinear systems. For these reasons, reinforcement learning (RL) has been proposed for the control and optimisation of energy systems including microgrids, energy storage and building energy management.
RL is a mathematical formalisation for learning-based control with the objective to learn a policy, through interaction in an environment, that maximises cumulative reward. Deployment of RL agents in the real world is sparse, and research is typically limited to domains that can be simulated to support cheap data generation required to train model-free deep RL algorithms. Model-based RL, which learn an explicit model of the environment dynamics, promise a number of advantages over model-free algorithms to make progress towards adoption in the real-world. These include improved sample efficiency reducing the number of environment interactions, greater scope for safety constraint satisfaction and enhanced explainability by offering insight into the predicted trajectory alongside the policy.
Collaborating with EDF R&D Digital Innovation, the aim of the PhD project is to make advancements towards the adoption of RL in the real world, with a focus on applications in the energy industry. My research focus is on developing safe model-based RL algorithms to meet explicit safety constraints, ensure trustworthy behaviour under environment uncertainty and enhance RL-human interaction to gain the confidence of end users.