Machine learning and physical-informed techniques for dynamic security assessment of changing grid dynamics

To decarbonize the electricity sector, an increasing number of renewables and energy storage are being integrated into electricity infrastructure. As a result, power grid dynamics are dominated not only by giant rotating machines, but also by semiconductor switches as well as their control algorithms. Dynamic security assessment (DSA) studies for this new paradigm can be based on either measurements or large interconnected grid simulations. Both studies result in big data challenges for DSA. Given a tremendous amount of energy management system (EMS) data (e.g., generator dynamics, network topology, and load models) and streaming synchronized measurements (e.g., synchrophasor and point-on-wave [POW] data), how do we assess the security of the power grids with such changing dynamics in a fast time scale? Performing model-based analysis on the EMS data for security assessment is not trivial, because present software cannot easily identify the core issues in the simulation, and it might not be capable to produce sufficient results for online analysis on time. Blackbox machine learning-based approaches to monitoring dynamic security may not be accepted by system operators, as their performance might not be interpreted by the first principles. This panel aims to discuss how to develop physically interpretable, and domain-tailored machine learning approaches to dynamic security assessment. It helps system operators deal with very large grid dynamic security assessment problems based on large-scale synchronized measurements and physical models. It will also connect experts from system operators (of both transmission and distribution systems), industry and academia to discuss implementation paths to overcome urgent threats to grid dynamic performance.