2. Inner Mongolia Regional Training Base of CAPF, Hohhot, 010070, China
Abstract: The classical Echo state network (ESN) cannot fully exploit its advantages in some work which characterize strong nonlinearity and high-order statistics. In order to overcome the shortcomings, this paper using the nonlinear readout nodes instead of the traditional linear readout nodes in classical ESN, Radical Basis Function (RBF) neural network was introduced to read out the reservoir state. Benchmark experiments on Lorenz chaotic time series and NARMA model identification shown that the proposed new type of ESN can improve the nonlinear characteristics and it’s performance exceeded the classical ESN in dealing with some higher degree of nonlinear system and model.
REFERENCES
[1] Jaeger H. The "echo state" approach to analyzing and training recurrent neural network[R]. GMD Report 148,
German Technical Report Centre for Information Technology, 2001.
[2] Jaeger H, Haas H. Harnessing nonlinearity: Predicting chaotic systems and saving energy in wireless
telecommunication[J]. Science, 2004, 304(5667): 78-80.
[3] Shi Zhiwei, Han Min. Support vector echo state machine for chaotic time-series prediction[J]. IEEE
Transactions on neural networks, 2007, 18(2): 359-372.
[4] Embrechts M, Alexandre L, Linton J. Reservoir computing for static pattern recognition[C]. 17th European
Symposium on Artificial Neural Network(ESANN 2009), Amsterdam, Netherlands, 2009, 245-250.
[5] Skowronski M D, Harris J G. Automatic speech recognition using a predictive echo state network classifier[J].
210 Neural Networks, 2007, 20(3): 414-423.
[6] Cernansky M, Tino P. Predictive modeling with echo state networks[C]. In Proceedings of the 18th International Conference on Artificial Neural Networks-ICANN, Prague, Czech Republic, September 3-6, 2008,778-787.
[7] Jaeger H. Adaptive nonlinear system identification with echo state networks[C]. In Advances in Neural
215 Information Processing Systems, 2002, 593-600.
[8] Venayagamoorthy G K. Online design of an echo state network based wide area monitor for a multi-machine power system[J]. Neural Networks, 2007, 20 (3): 404-413.
[9] Hartland C, Bredeche N, Sebag M. Memory-enhanced evolutionary robotics, The echo state network approach [C]. IEEE Congress on Evolutionary Computation, Trondheim, Norway, May 18-21, 2009, 2788-2795.
[10]Mantas Lukoˇseviˇcius. Echo State Networks with Trained Feedbacks, Technical Report No. 4, February 2007.
[11] Lukosevicius M, Jaeger H. Reservoir computing approaches to recurrent neural network training[J]. Computer Science Review, 2009, 3(3): 127-149.
[12] Levy Boccato, Diogo C. Soriano, Romis Attux and Fernando Jos´e Von Zuben. Performance Analysis of Nonlinear Echo State Network Readouts in Signal Processing Tasks. WCCI 2012 IEEE World Congress on Computational Intelligence June, 10-15, 2012 - Brisbane, Australia.
[13] ZHANG Dong-qing, NING Xuan-xi, LIU Xue-ni. On-line prediction of nonlinear time series using RBF neural networks. Control Theory & Applications, Vol. 26 No. 2 Feb. 2009:151-155.
[14] S. H. Strogatz, Nonlinear Dynamics and Chaos: With Applications to Physics, Biology, Chemistry and Engineering. Westview Press, 2000.
[15] PANG Shi-wei, YU Kai-ping, ZOU Jing-xiang. “Identifying nonlinear time-varying structural system based on NARMA model”. JOURNAL OF HARBIN INSTITUTE OF TECHNOLOGY, Vol.40 No 1,Jan. 2008:12-16.
作者簡介：
烏日娜，（1982-），碩士研究生，主要研究方向：通信系統智能信號處理。
