A Multiple Randomized Learning based Ensemble Model for Power System Dynamic Security Assessment

Abstract

With the deployment of wide-area measurement systems (WAMS), intelligent learning techniques have recently shown their strengths in online power system dynamic security assessment (DSA). This paper proposes a multiple randomized learning based ensemble model for online DSA. Rather than using a single learning algorithm, the proposed model combines multiple randomized learning algorithms, including extreme learning machine (ELM) and random vector functional link networks (RVFL), to obtain more diversified machine learning outcome. Based on such diversified outputs, a credible decision-making process is designed to optimally discriminate credible and incredible DSA results, so a high DSA accuracy can be ensured with a high DSA efficiency. The proposed model is tested on New England 39-bus system, which demonstrates its improved DSA performance over a single learning algorithm.