Multivariate Time Series Prediction Based on a Simple RBF Network

Abstract

Improvement on defining hidden layer’s clustering centers of RBF network is greatly important for the network modeling and prediction, especially for multivariate time series. This paper firstly uses a linear function and a nonlinear function respectively to detect the linear correlations and the nonlinear correlations of the time series. And a small data set which includes effective information of the system is defined. Then, a local search procedure is introduced to optimize K-means clustering algorithm, which is aimed at quickly and effectively adjusting hidden layer’s clustering centers. Finally, input other training samples to determine network weights, LS or its improved method can be selected. Simulation results show that compared with that of conventional RBF network, with the same number of hidden layer’s nodes, the RBF network model in this paper gets better prediction accuracy.

Nonlinear System Identification Based on an Online SCFNN With Applications in IoTs

“…When the prediction error is 0, E PA =1. In summary, E RMSE indicates the absolute deviation between prediction and observation value, and E PA reacts the similarity between prediction and observation data (Xi & Wang, 2012).…”

“…In the second example, the governing equation of the system is given by (Narendra & Parthasarathy, 1990;Xi & Wang, 2012), that is where the output at time (n+1) is a nonlinear function of the outputs at times (n) and (n -1) plus a linear function of the input at time (n). The reference input u(n) is selected as u(n) = sin(2πn/25).…”

“…In the third example, the governing equation of the system is given by (Narendra & Parthasarathy, 1990;Xi & Wang, 2012), that is 4 shows the simulation results numerically.…”