OPTIMASI LEARNING RADIAL BASIS FUNCTION NEURAL NETWORK DENGAN EXTENDED KALMAN FILTER

Oni Soesanto, Arfan Eko Fahrudin, Dodon T. Nugrahadi

Abstract


Dalam paper ini dibahas mengenai optimasi Radial Basis Function Neural Network (RBFNN) dengan Extended Kalman Filter. Proses learning RBF dengan Extended Kalman Filter menggunakan parameter bobot pada hidden center RBF yaitu noise proses pada perhitungan bobot hidden center dan noise pengukuran pada data output. Extended Kalman Filter pada jaringan syaraf RBF berfungsi mengoptimalkan bobot pada hidden center dengan meminimalkan error pada output RBF dengan parameter proses pada unit center RBF dan parameter bobot output pada output layer. Bobot output optimal diperoleh pada saat error output pada training RBF telah konvergen, selanjutnya digunakan untuk proses testing. Algoritma Extended Kalman Filter dan Radial Basis Fuction (EKF-RBF) memungkinkan proses learning memungkinkan center dan variansi pada hidden layer tidak perlu dihitung sebelum bobot output optimum ditemukan. Hasil simulasi menunjukkan bahwa pada training, performansi klasifikasi algoritma EKF-RBF mampu mengenali rata-rata 92.42% dan untuk prediksi didapatkan MAE sebesar 5,3846 dan RMSE sebesar 16,2398 dengan CPU time 24,4146 detik dengan iterasi rata-rata 68,8 iterasi, testing in sample rata-rata MAE sebesar 4,3388, rata-rata RMSE sebesar 13,2230 dan rata-rata CPU time sebesar 0,1123 detik sedangkan pada testing out sample didapatkan rata-rata MAE sebesar 4,1065, RMSE sebesar 11,0126 dan CPU time sebesar 0,0265 detik.

Kata kunci : Extended Kalman Filter, Extended Kalman Filter – Radial Basis Function (EKF-RBF), Optimasi Jaringan Syaraf RBF

Full Text:

PDF

References


Amir, Panah, 2013, Enhanced SLAM for a Mobile Robot using Unscented Kalman Filter and Radial Basis Function Neural Network, Research Journal of Recent Sciences, Vol.2(2), 69-75

Cho, B.H., Yu, H., Lee, J. dan Kim, I.Y. 2008.Nonlinear Support Vector Machine Visualization for Risk Factor Analysis Using Nomograms and Localized Radial Basis Function Kernels. IEEE Transaction On Information Technology in Biomedicin.12: 247-256.

de Oliveira, Mauri Aparecido, 2012, An Application of Neural Networks Trained with Kalman Filter variants (EKF and UKF) to Heteroscedastic time series Forecasting, Applied Mathematical Sciences, Vol. 6, No. 74, 3675-3686

Dachapak, C., Kanae, S., Yang, Z. J., & Wada, K. 2004. Orthogonal least squares for radial basis function network in reproducing kernel hilbert space, hlm.847-848. IFAC Workshop On Adaptation and Learning in Control and Signal Processing, and IFAC Workshop on Periodik Control System. Yokohama, Japan.

Fausett, L. 1994.Fundamentals of Neural Networks: Architectures, Algorithms and Applications. Prentice-Hall, New Jersey, USA.

Gupta Madam M., Liang Jin dan Noriyasu Homma, 2003. Static and Dynamic Neural Networks: From Fundamentals to Advanced Theory, John Wiley & Sons, Inc.

Haykin, S., 2001. Kalman Filtering and Neural Network, John Wiley & Sons, Inc.

Lu, W.Z., Wang, W.J., Wang, X.K., Yan, S.Y. dan Lam, J.C. 2004. Potensial Assesment of A Neural Network Model with PCA/RBF Approach Forecasting Pollutant Trends in Mong Kok Urban Air, Hong Kong’, Environmental Research 96, 79–87.

Liu, X., Guobin, C., dan Baiqing, H. 2012. Neural Network Forecast Algorithm Based on Iterated Unscented Kalman Filter, The 2nd International Conference on Computer Application and System Modeling, Antlantis Press, Paris.

Simon, Dan, 2002, Training Radial Basis Neural Network with the Extended Kalman Filter, Neurocomputing, Vol. 48. Issues 1-4, 455-475

Trebaticky, P. 2005. Recurrent Neural Network Training with the Extended Klaman Filter, M. Bielikova (ed), IIT.SCR, 57-64

Wang, X.Z., Li, C.G., Yeung, D.S., Song, SJ dan Feng, H.M. 2008. A Definition of Partial Derivative of Random Functions and Its Application to RBFNN Sensitivit Anaysis, Neurocomputing, 71, 1551–1526.

Welch, G. dan Bishop, G., 2001. An Introduction to the Kalman Filter, University North Carolina.

Yousef R. dan Hindi, K. 2006. Locating center point for radial basis function network using instance reduction techniques. World Academy of Science, Engineering, and Technology.4: 213-216.




DOI: http://dx.doi.org/10.20527/klik.v2i2.40

Copyright (c) 2016 KLIK - JURNAL ILMIAH ILMU KOMPUTER



Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.joomla
counter View My Stats