Nonlinear system identification using delta learning-based gaussian-hopfield networks

Abstract

本論文針對非線性離散時間系統，提出一種以最徒負梯度法則為基礎之高斯-霍普菲爾類神經網路之新驗證方法。並藉由兩種不同的離散時間模型來描述此單輸入單輸出之非線性離散時間系統。而此等效非線性系統模型之非線性部份可用以相關聯之輸入及輸出組成的非線性函數來等效之。我們用高斯基組函數去代表? 一個非線性離散時間系統模型之非線性函數，以最徒負梯度法則去訓練一組高斯基組函數之係數並使其最佳化。最後以模擬結果來驗證此方法之逼近效能。

This study proposes a new verifying method of Gaussian-Hopfield neural networks (GHNNs) based on Delta-learning rules in the light of the discrete-time nonlinear system, and describes the discrete-time nonlinear system of Single-Input-Single-Output (SISO) with two different discrete-time models. The nonlinear part of the equivalent nonlinear system can be equaled with related input and output nonlinear functions. In the paper, Gaussian basis functions (GBFs) was used to represent the nonlinear function of a discrete-time nonlinear system and optimize the GBFs by training it with Delta-learning rules. Finally, the simulation results were used to verify the approach effect.