Machine Learning and Modern Heuristic Optimization for Planning and Operation of Active Distribution Networks

As a result of the continuous digitalization pace of the energy sector, larger amounts of data streams are available now. From the distribution system operators (DSOs) perspective, the deployment of ICT infrastructure enables real-time monitoring and optimal management of assets. At the same time, the increasing introduction of Internet of Things (IoT)-based devices allows users to characterize and adapt their energy consumption behavior. At the same time, the increasing number of distributed energy resources (DERs) on distribution networks demands DSOs and users a better level of coordination, aiming at continuously guaranteeing reliable and cost-effective operations. In this context, innovative approaches based on machine learning (ML), capable of learning from all the data available, can guarantee such an expected level of coordination. Nevertheless, the current ML method lacks reproducibility of results, while others lack convergence guarantee if the data is not properly curated. To address this, modern optimization methods (e.g., classical mathematical programming, heuristic optimization) can be used to enhance AI-based methodsí performance. In this panel, we gather academics and industry experts to discuss initiatives concerning the application of modern optimization and ML methods in active distribution networks. Thinking about the future adoption of such optimization-AI-based approaches, technical challenges, such as convergence guarantee, data efficiency, computational efficiency, robustness, explainability, and new concerns related to trust, safety, privacy, fairness, etc., will be discussed. Presentations in this panel session: - Combing Mathematical Programming and Machine Learning to Enforce Safety in Active Distribution Networks Operation (23PESGM1161) - On the importance of receptive field in graph neural network applications for power systems (23PESGM1163)