A power system stability assessment framework using machine-learning

With the planned massive integration of renewable energy resources, the electric network is gradually becoming a less predictable system. In such power systems, the set of critical operating points shifts and no longer necessarily coincides with peak and light load conditions. This means that deducing the stability of the system from a restricted analysis of a limited number of operating points becomes very speculative. A methodical approach that covers all hourly operating points throughout a study year is therefore necessary. This paper offers a platform based on supervised and unsupervised machine learning techniques along with optimal power flow, steady-state, and dynamic simulation tools to perform deterministic time-series simulations. The platform allows rapid time-series preliminary assessments of transient stability (i.e., rotor angle stability) in the context of high renewable penetration. As a case study, the proposed platform is tested to assess the transient stability of an IEEE-39 test system augmented with renewable energy resources. Various generation expansion scenarios are considered to show how the platform can be used to conduct grid scenario analyses by investigating the effect of different energy mixes, renewable penetration levels, and renewable siting.