Learning for Power Distribution System Optimization, Control, and Protection

This panel discusses recent and exciting ideas on how to exploit data and leverage machine learning to aid computationally challenging grid operation tasks. Advances in numerical optimization and commercial solvers have allowed distribution grid operators to handle a fairly broad class of operations. However, the rampant integration of renewables call for the need to solve optimal power flow (OPF), battery placement and sizing, and stable control or adaptive protection setting tasks at increasingly faster timescales and over uncertain input parameters. When complexity can hinder real-time operation, learning from big data can be a viable alternative or at least aid in creative ways to accelerate conventional alternatives. In particular, renewable generation and demand together with their OPF decisions can be used to train neural networks to expedite algorithms. Moreover, novel methods from reinforcement learning can promptly optimize inverter settings under the network-constrained and multi-agent setting found in distribution grid operation. Challenging chance-constrained OPF formulations can be dealt with DNN-based approaches to design optimal inverter control policies. Adaptive protection schemes are designed by genuine approaches from innovations learning. Presentations in this panel session: - Controlling Power Distribution Systems with Safe Reinforcement Learning (23PESGM2742) - Controlling Smart Inverters using Proxies: A Chance-Constrained DNN-based Approach (23PESGM2747)