An Investigation on Deep Learning and Multi-label Learning for Composite System Reliability Evaluation

Abstract

In many cases, research on reliability analysis focuses on searching the state space of the system for states that represent events of interest, like failure of the system not meeting the required demand for a specific node. This raises the need for search procedures that efficiently determine states to be examined and then evaluated. Artificial Intelligence based methods have been studied for this objective either by themselves or in conjunction with widely used methods like Monte Carlo Simulation. This dissertation investigates various novel approaches for reliability evaluation of composite power systems by combining Monte Carlo simulation (MCS) with different machine learning techniques for Multi-Label Learning and Deep Learning topologies. The objective in this research is reducing the computational burden to perform Monte Carlo Simulation for a given level of accuracy. As a consequence, higher accuracy can be obtained for the same level of computational effort