An Artificial Neural Network Based Strategy for Commutation Failure Forecasting in LCC-HVDC Transmission Networks

Commutation failure (CF) is a major challenge associated with line-commutated converter high voltage dc (LCC-HVdc) systems. Faults at the inverter ac systems are regarded as main cause of commutation failure. The present study proposes a method for forecasting the commutation failure based on a feed-forward neural network (FFNN) combined with optimization algorithms and principal component analysis (PCA). To fulfill this, first, a two-terminal HVdc system is designed, and then, a large number of three-phase faults with different fault inductances are applied on the inverter ac side. In each fault case, 100 samples of the dc current signal in the interval between fault and commutation failure initial times are taken with the sampling frequency of 1 MHz. Subsequently, with the aim of dimensionality reduction, the samples are preprocessed by the PCA before serving as inputs of the neural network. Finally, various metaheuristic optimization methods, i.e. particle swarm optimization (PSO), grey wolf optimizer (GWO), simulated annealing (SA), and genetic algorithm (GA) are utilized to determine the most appropriate structure for the neural network in terms of the number of neurons in each hidden layer, and the type of transfer function (TF) in hidden layers and output layer. The simulation results indicate that the optimized artificial neural network can predict the commutation failure occurrence with an accuracy of roughly 90%.