Comparing performances of three CFNN used for DC machine combined parameter and states estimation

Abstract

The main idea of this paper is for evaluation and make a performance comparison between three types of learning algorithms in the case of simultaneous estimation of parameter and states of a brushed DC machine. Three Cascade Forward Neural Network (CFNN) estimators have been designed, the first one is based on Quasi-Newton BFGS backpropagation (BFGSBP), the second one is based on Resilient backpropagation (RBP) and the last one is based on Bayesian Regularization backpropagation (BRBP). All this neural network use just voltage and current as imputes and estimates simultaneously speed, temperature and armature resistance. A series of simulation have been carried out for three algorithms and the results were compared between them for each Artificial Neural Network (ANN) outputs. The comparative study of the time required to converge for each supposed MSE, present the trade-off between fastness and convergence of three algorithms in order to develop the best NN.