Title :
Parameters selection method of the leaky integrator echo state network for time series
Author :
Xianshuang Yao ; Shuxian Lun
Author_Institution :
Coll. of Eng., Bohai Univ. Jinzhou, Jinzhou, China
Abstract :
In this paper, we present a relation of the parameters selection of the Echo State Network (ESN), thus we can obtain the optimal performance of ESN. The parameters include mainly the leaking rate (μ), the spectral radius (μμ) and the size (μ) of reservoir units. For the limited length of a training sequence, a formulation of ESN parameters is developed to improve the training and testing accuracy and efficiency of ESN. Simulation experiments demonstrate that the formulation is valid.
Keywords :
mathematics computing; recurrent neural nets; time series; ESN; echo state network; leaking rate; parameter selection method; reservoir unit size; spectral radius; time series; training sequence; Autoregressive processes; Data models; Mathematical model; Recurrent neural networks; Reservoirs; Testing; Training;
Conference_Titel :
Mechatronics and Control (ICMC), 2014 International Conference on
Print_ISBN :
978-1-4799-2537-7
DOI :
10.1109/ICMC.2014.7231760