Machine-learning-based reliability evaluation framework for power distribution networks

Abstract

Reliability evaluation of power distribution systems is a well-studied and understood problem, but topology analysis efficiency and failure feature description are still limiting the practical application of reliability evaluation in distribution systems. One approach the authors proposed in this study is to incorporate the machine-learning technique into the simulation-based reliability evaluation method, and assess the system reliability in an empirical fashion. In this study, a framework for performing the machine-learning-based reliability evaluation, and corresponding modelling processes are first established. Then, by inducing a supervised learning algorithm named perceptron, a state-space classification-based (SSC) method for system state assessment, the core procedure of reliability evaluation, is proposed. On this basis, a reliability evaluation algorithm combining the SSC-based system state assessment and sequential Monte Carlo simulation is proposed, where the workload of topology analysis required in conventional reliability evaluation methods can be released. Furthermore, extensive case studies are conducted on the Roy Billinton Test System (RBTS) bus 2 system and an actual distribution system to verify the proposed models and algorithms. Results show that the proposed framework supports a more efficient reliability evaluation pattern while ensuring the evaluation accuracy.