SCADA-informed surrogate modeling and evolutionary optimization for simultaneous exergy, cost, and vibration management in gas turbine systems

The transition toward sustainable industrial power generation requires highly efficient and reliable energy conversion systems. However, balancing thermodynamic performance with economic viability and mechanical integrity under fluctuating operational loads remains a critical engineering challenge. Traditional optimization frameworks typically rely on computationally intensive equation solvers and frequently neglect real-world mechanical degradation, thereby limiting their utility for dynamic control. To address these limitations, this study proposes a prescriptive analytics framework that integrates physics-based thermodynamic modeling with data-driven surrogate algorithms to optimize gas turbine operations. Using real-world operational data, Gradient Boosting, Artificial Neural Network, and Random Forest models were trained and benchmarked to accurately predict exergy efficiency, total cost rates, and maximum vibration. The highest-performing models were subsequently employed as objective functions within a Non-dominated Sorting Genetic Algorithm II to generate multidimensional Pareto frontiers. A multi-criteria decision-making method was then applied to isolate the optimal operating setpoint. The prescribed tri-objective solution enhanced overall exergy efficiency by 6.0 percent compared to the historical baseline, simultaneously reducing operational cost rates to 60.47 $/h and constraining maximum mechanical vibration to 24.856 μm. Ultimately, this hybrid multi-criteria methodology provides plant operators with actionable, real-time control parameters that harmonize performance, cost, and reliability, ensuring long-term systemic sustainability.