Decentralized Energy Management of Microgrids: When Optimization Meets MachineLearning

International audienceDecarbonization calls for a deep transformation of energy systems. This mutation is supported by an increase of renewables penetration in the energy mix. Though, as they are intermittent and non-dispatchable, renewable energy sources pose various challenges. Therefore, the future grid is expected to be an interconnected network of both small-scale microgrids and large-scale developments.Microgrid is defined as small-scale localized power grid which can either work singly or compete with nearby main electrical grid. The emerge of this new subsystem with local information and means of action leads to numerous energy management challenges. In this new energy ecosystem with distributed renewable energy, ubiquitous data and computation power, optimization alone is no longer sufficient. This calls to supplementary techniques. For instance, machine learning offers the possibility to automate the data gathering, modelling, and optimization stages. In this framework, optimization can be coupled to machine learning as they answer to different objectives.In this paper we investigate more deeply the challenges of microgrids management. Existing literature uses numerous approaches, focusing on smart grid and microgrid specific management issues. Though, to our best of knowledge, no overview aligning optimization and machine learning exists. This paper contributes to the understanding of the joint role of both cited techniques in tackling decentralized energy management of microgrids.‣Optimization targets one problem class, or even an instance of a problem, and a theory geared towards optimality and efficiency;‣Machine learning is focused on statistical significance, replicability to other problems with few adaptations, and interpretability of results