Neural network modeling for rigorous simulation of mid-pressure reciprocating expansion engines performance

Abstract

In the Natural Gas (NG) distribution network, the town border stations are used to reduce NG pressure. Currently, pressure reduction is done by using throttling valves that cause a significant amount of physical exergy loss. Reciprocating Expansion Engine (REE) is a new tool to recover this pressure exergy. This study presents a thermodynamic evaluation to simulate the REE rigorously. The GERG-2008 equation of state (EOS) as an accurate model is used to calculate NG thermodynamic properties. As the utilization of GERG-2008 EOS requires specific inputs (pressure, temperature and mole fraction), the Try & Error method is utilized (while the inputs are density, specific internal energy, and NG composition). An Artificial Neural Network (ANN) method is also developed as an additional and alternative tool to overcome a basic limitation of the Try & Error method (knowing the NG composition). The validation results show that the error of the ANN and Try & Error methods for temperature, pressure calculation are reported 0.03%, 0.08%, and 0.04%, 0.9%, respectively. Also, the results show that the NG compositions have a significant impact on REE performance so that the indicated work for Khangiran and Bidboland NG mixtures are reported 165 and 143 kJ/kg, respectively.