Design of an ensemble neural network to improve the identification performance of a gas sweetening plant using the negative correlation learning and genetic algorithm

This paper presents a combination of negative correlation learning (NCL) and Genetic Algorithm (GA) to create an ensemble neural network (ENN). In this approach the component neural networks (CNNs) of ENN are trained simultaneously. The resulting CNNs negatively correlate together through the penalty terms in their objective functions. The predicted output is obtained by using the weighted averaging of the outputs of CNNs. GA participates in the training of CNNs and assigns proper weights to each trained CNN in the ensemble. The proposed method was tested on a case study in the Gas Treatment Plant (GTP) of the AMMAK project in the Ahwaz onshore field in Iran. The testing results of the model properly follow the experimental data. In addition, the proposed method outperformed the single neural network and some other network ensemble techniques.