An Ensemble of Neural Networks for Control System Application

DOI: 10.5176/2251-2179_ATAI12.30

Authors: Alvin Sahroni

Abstract: This study shows that the convergence towards a good performance do not have to rely large architecture. Generally, Backpropagation Artificial Neural Network (BPANN) method was used in control system applications. One of the purposes related to supervised learning performance and ease of implementation. In control system, the training of BPANN should be based on its Minimum Square Error (MSE) to get a good transient response within its performance. This study uses 3 Node of Neurons within Hidden Layers, and using 3 BPANN that will be ensemble. By using an ensemble application of BPANN within temperature control system, it was found that the "generalization" error can be derived and give result about 95{6e6090cdd558c53a8bc18225ef4499fead9160abd3419ad4f137e902b483c465} of performance. Based on the result, it was conclude that ensemble method in control system applications is very reliable and have a good prospective for future investigation.

Keywords: BPANN, Ensemble, generalization error, Backpropagation, Neural Networks