A Data-driven Methodology for Transient Stability Assessment Based on Broad Learning System

Abstract

This paper proposes a data-driven methodology for transient stability assessment (TSA) based on constructing a transient stability boundary (TSB). Without stacking the network layers, the TSB construction algorithm makes a broad expansion in the neural nodes thereby forming a clear structure that can be theoretically analysed. While preserving a high accuracy and generalization ability, the TSB expression is clear, differentiable and therefore applicable to dynamic security constrained problems. Furthermore, a transfer learning strategy (TLS) is employed to build TSBs from a limited number of samples in a time-saving way. The possibilities of the developed method are tested via case study that uses the IEEE 39-bus test system. The case study confirmed that the introduced algorithm is highly precise and insensitive to the number of available samples/parameters. This indicates that the proposed method is effective, robust and that as such it may serve as a valuable tool of online TSA.