Title

Prediction of water removal rate in a natural gas dehydration system using radial basis function neural network

Document Type

Article

Publication details

Tatar, A, Nasery, S, Bahadori, M, Bahadori, A, Bahadori, M, Barati-Harooni, A & Najafi-Marghmaleki, A 2016, 'Prediction of water removal rate in a natural gas dehydration system using radial basis function neural network', Petroleum Science and Technology, vol. 34, no. 10, pp. 951-960.

Published version available from:

http://dx.doi.org/10.1080/10916466.2016.1166131

Peer Reviewed

Peer-Reviewed

Abstract

Natural gas commonly contains water as a contaminant that can condense to water or form gas hydrates, which causes a range of problems during gas production, transportation, and processing. Therefore, the removal of gas moisture is of great importance. A common and popular method for removing water contamination from natural gas is using solid dehydrators. Calcium chloride is a nonregenerative desiccant to dehydrate natural gas. With continual water adsorption, CaCl2 changes to consecutively higher states of hydration, finally producing a CaCl2 brine solution. This method does not require heating or moving parts. In addition, it does not react with H2S or CO2. These features make this method a popular one for drying natural gas. Nevertheless, precise and simple methods are needed to predict the water content of natural gas dried by calcium chloride dehydrator units. In this study, an intelligent method, called the radial basis function neural network, was incorporated to predict the gas moisture dehydrated by calcium chloride in dehydration units. Modeling was performed under different conditions of a fresh recharge and before recharging. The overall correlation factor of 0.9999 for both the fresh charge and before charging conditions showed that the outputs of the proposed models were in agreement with the experimental data. In addition, the developed models were compared with the previously proposed intelligent models and classic correlations. The comparison showed that the developed model is superior to the previously proposed models and correlations regarding the accuracy of prediction.