Split-based sequential sampling for realtime security assessment

Machine learning (ML) for real-time security assessment requires a diverse training database to be accurate scenarios beyond historical records. Generating operating conditions is highly relevant the uncertain future of emerging power systems that are completely different systems. In response, first time, this work proposes novel split-based sequential sampling approach based on optimisation generates more operation ML models than state-of-the-art approaches. This also volume-based coverage metric, convex hull volume (CHV), quantify quality samplers generated data. metric accounts distribution samples across multidimensional space measure within physical network limits. Studies IEEE test cases with 6, 68 and 118 buses demonstrate efficiency approach. Samples using proposed cover 37.5% random in 68-bus system. The CHV can assess (standard deviation 0.74) better distance-based which outputs same value <0.001) very data distributions As we demonstrate, as pre-step critical tasks such assessment.