Multilevel inverter fault diagnosis using optimised radial basis neural network — A novel performance enhancement

Abstract

This paper is an attempt to develop a new technology, which is an advancement of the previously published paper [1] for fault diagnosis of multilevel inverter adopting the machine learning and optimization techniques. The advanced machine-learning algorithm called the Optimized Radial Basis Neural Network (ORBNN) method is developed in which the Neural Network uses Radial Basis function as the activation function. So in this paper the faulty condition of switch is identified using the neural network. Matlab based implementation is carried out using the neural network and means square error is minimized by using radial basis function neural network, trained with parameter optimization techniques gives better results. The fault diagnosis is carried out on a 7-level cascaded H-bridge inverter using neural network trained with Back-Propagation (BP), and optimized using Genetic Algorithm (GA), Particle Swarm Optimization (PSO) and Cuckoo Search Algorithm (CSA). The parameter chosen for the optimization is Mean Square Error(MSE), which has been minimized for the neural network and the radial basis neural network using different optimization algorithm as mentioned above. Matlab based simulation inferred that radial basis neural network performed better than the ordinary neural network and the CSA optimized radial basis neural network (ORBNN) delivered the lowest MSE concluding itself as the best method among the methods taken for analysis.