Novel Gas-oil Separation Plant Design in Mediterranean and Synthetic Data Generation for Machine Learning Model

This study presents the development of a Gas Oil Separation Plant (GOSP) simulation using ASPEN HYSYS and the Peng Robinson equation of state to model the thermodynamic behavior of hydrocarbon mixtures. The design was tested under a range of feed conditions, including varying gas and condensate flow rates, to improve separation and processing efficiency. Field data from the Mediterranean concession, including reservoir properties, well characteristics, and fluid compositions, formed the basis of the model. Key parameters such as temperature, pressure, gas to oil ratio, and the mole fractions of methane and ethane were considered. Fluid characterization was carried out using PVT modeling software by Schlumberger. A Python script was developed to automate HYSYS simulations via the "win32com" interface, generating a synthetic dataset of 1,700 operational scenarios. Three experimental cases explored the effects of separator pressures, compressor limitations, and Reid Vapor Pressure on system performance. The resulting synthetic data achieved over 91 percent agreement with simulation outputs using linear regression proxy model, while terminal gas heating value reached 9691 kcal per cubic meter and condensate Reid Vapor Pressure remained within the 82.7 kPa limit. These outcomes confirm the model’s accuracy and its potential for supporting design optimization and operational decision-making.