Performance comparison between HMLP, MLP and RBF networks with application to on-line system identification

This paper compares the performance of hybrid multilayered perceptron (HMLP), multilayered perceptron (MLP) and radial basis function (RBF) networks. These networks were tested to perform online system identification of nonlinear systems. Two sets of data were used for this comparison, one simulated data set and one real data set. The results for both data sets indicated that HMLP network gave significant improvement over standard MLP network. The additional linear input connections of HMLP network do not significantly increase the complexity of MLP network since the connections are linear. In fact by using the linear input connections, the number of hidden nodes required by the standard MLP network model can be reduced that would also reduce computational load. It was also found that HMLP network gave better performance and more efficient than RBF network. HMLP network has less adjustable parameters but could offer better performance than RBF network.