Modified coordinate descent methodology for solving process design optimization problems: Application to natural gas plant

• Modified coordinate descent methodology is proposed for NG plant optimization.
• Coordinate descent with random search and box search are the main highlights.
• Application of MCD for KSMR NG liquefaction cycle is presented.
• Simple implementation, robust convergence makes MCD practical and useful.

A modified form of Coordinate Descent methodology (MCD) is presented for solving process optimization problems. The modifications made to the conventional coordinate descent algorithm include search initialization inspired by a pattern search, sequential coordinate randomization for exploring search space and box search for refining of local optimum. The performance of the proposed MCD methodology was examined on benchmark mathematical test problems. After successful convergence of the test problems in reasonable time the MCD algorithm was exploited for the optimization of Natural gas (NG) liquefaction process plant developed in a commercial simulator. The newly developed Korea Single Mixed Refrigerant (KSMR) process was optimized for compression energy demand which is a strong function of refrigerant composition and its operating pressures. MCD was successful in finding the optimum refrigerant composition and operating pressures levels that results in energy savings of 40% and 11% compared with the representative base cases. The suitability of MCD algorithm for NG process plant was further demonstrated by comparing the results of KSMR process with PSO and NSGA-II algorithm. The comparison results demonstrate a nominal improvement in terms of energy savings however the calculation time and ease of implementation and independence of MCD on parameters give it clear advantage. Thus is suitable for solving process design optimization problems particularly related NG plant.