Algoritmo de entrenamiento óptimo para diseñar una memoria asociativa de diagnóstico de fallas

José A. Ruz Hernández, Edgar N. Sánchez, Dionisio A. Suárez

Resumen

En este artículo, los autores presentan un nuevo enfoque de síntesis para entrenar memorias asociativas implementadas con redes neuronales recurrentes. Los pesos de la red recurrente se determinan como la solución óptima de la combinación lineal de vectores soporte. El algoritmo de entrenamiento propuesto maximiza el margen entre los patrones de entrenamiento y la superficie de decisión. El problema de diseño considera: 1) la obtención de los pesos por medio del algoritmo de hiperplano óptimo utilizado para máquinas de vector soporte y 2) la obtención de las condiciones para reducir el número total de memorias espurias. El nuevo algoritmo desarrollado se utiliza para diseñar una memoria asociativa que diagnostique fallas en centrales termoeléctricas.

Palabras clave

Memoria asociativa; red neuronal recurrente; máquinas de vector soporte; hiperplano óptimo; detección y diagnóstico de fallas; central termoeléctrica

Texto completo:

pdf

Referencias

Boser, B. E., E. M. Gullon and V. N. Vapnik (1992). A training algorithm for optimal margin classifiers. Proceedings of the Fifth Annual Workshop of Computational Learning Theory 5, 144–152.

Chartier, S. and R. Proulx (2006). Ndram: Nonlinear dinamic recurrent associative memory for learning bipolar and nonbipolar correlated patterns. IEEE Transactions on Neural Networks 16, 1393–1400.

Cortes, C. and V. N. Vapnik (1995). Support vector networks. Machine Learning 20, 273–297.

Hassoun, M. H. (1993). Associative Neural Memories: Theory and Implementation. Oxford Univ. Press. Oxford, U. K.

Hassoun, M. H. and A. M. Youssef (1990). Asociative neural memory capacity and dynamics. Proceedings of International Joint Conference on Neural Networks 1, 763–769.

Haykin, S. (1999). Neural Networks, A comprehensive foundation. Prentice Hall. New Yersey, USA.

Kuhn, H. W. and A. W. Tucker (1961). Nonlinear programming. Proceedings of the Second Berkeley Symposium on Mathematical, Statistics and Probability pp. 481–492.

Lee, D. L. and T. C. Chuang (2006). Designing asymetric hopfield-type associative memory with higher order hamming stability. IEEE Transactions. on Neural Networks 16, 1464–1476.

Li, J. H., A. N. Michel and W. Porod (1988). Qualitative analysis and synthesis of a class of neural networks. IEEE Transactions on Circuits and Systems CAS-35, 976–986.

Liu, D. and A. N. Michel (1994). Dynamical Systems with Saturations Nonlinearities: Analysis and Design. Springer Verlag. New York, USA.

Liu, D. and A. N. Michel (1996). Robustness analysis and design of a class of neural networks with sparse interconnecting structure. Neurocomputing 12, 59–76.

Liu, D. and Z. Lu (1997). A new synthesis spproach for feedback neural netwoks based on the perceptron training algorithm. IEEE Transactions on Neural Networks 8, 1468–1482.

Norgaard, M. (2000). Neural Networks Based System Identification Toolbox for Use with MATLAB, Technical Report 00-E-891. Deparment of Automation, Technical University of Denmark. New York, USA.

Ruz-Hernandez, J. A. (2006). Development and Application of a Neural Network-based Scheme for Fault Diagnosis in Fossil Electric Power Plants, Ph. D. Thesis. Centro de Investigacion y Estudios Avanzados del Instituto Politecnico Nacional (CINVESTAV-IPN). Guadalajara Campus (in Spanish).

Ruz-Hernandez, J. A., E. N. Sanchez and D. A. Suarez (2007). Optimal Training for Associative Memories: Application to Fault Diagnosis in Fossil Electric Power Plants. Book Chapter of Hybrid Intelligent Systems, Analysis and Design, Edited by O. Castillo et al., International Series in Fuzzyness Studies and Soft Computing, Vol. 208, pp. 326-359, Springer-Verlag-Heidelberg. Germany.

Sanchez, E. N. and A. Y. Alanis (2006). Neural Networks, Foundations and Applications to Automatic Control. Pearson Education. Madrid, España (in Spanish).

Vapnik, V. N. (1982). Estimation of Dependences Based in Empirical Data, Addendum I. Springer Verlag. New York, USA

Abstract Views

493
Metrics Loading ...

Metrics powered by PLOS ALM




Creative Commons License

Esta revista se publica bajo una Licencia Creative Commons Attribution-NonCommercial-CompartirIgual 4.0 International (CC BY-NC-SA 4.0)

Universitat Politècnica de València     https://doi.org/10.4995/riai

e-ISSN: 1697-7920     ISSN: 1697-7912