Restricted Boltzmann Machines Based Fault Estimation in Multi Terminal HVDC Transmission System

Abstract

The facilitation of bulk power transmission and non-synchronized interconnection of alternating current (AC) grids convince engineers and researchers to explore high voltage direct current (HVDC) transmission system in a comprehensive way. This exploration focuses on control and protection of HVDC transmission system. Fault estimation is a core component of protection of HVDC transmission system. This is because of sudden built up of direct current (DC) fault. In this research, DC fault is estimated in multi terminal HVDC transmission system based on restricted Boltzmann machine. Restricted Boltzmann machine is a generative stochastic artificial neural network in which learning of probability distribution is conducted over the set of inputs. Three terminal HVDC transmission system is simulated under normal and faulty conditions to analyze variations in electrical parameters. These variations serve as learning parameters of restricted Boltzmann machine. Contrastive divergence algorithm is developed to train restricted Boltzmann machine. It is an approximate maximum likelihood learning algorithm in which gradient of difference of divergences is followed. It is found that fault is estimated with the testing of variations in minimum time steps. Simulation environment is built in Matlab/Simulink.