Digital twin-enabled dynamic 4E-S multi-objective optimization framework for intelligent gas turbine operation with flexible loads

Mature gas turbines effectively address challenges from renewable intermittency to grid reliability. However, current gas turbine flexible-load operating strategies often neglect dynamic environmental factors and performance degradation's impact on the 4E-S (energy, exergy economics, emissions, and sustainability) framework, potentially compromising energy-saving and emissions reduction targets. Rooted in digital twin, this study presents a comprehensive lifecycle framework for intelligent, optimal gas turbine operation. Initially, a source domain hybrid model is developed. Subsequently, leveraging transfer learning principles, a hybrid transfer learning methodology is implemented for modeling the target domain. The gas turbine operating window is then constructed, and a thorough 4E-S analysis is performed. Further, a novel dynamic multi-objective evolutionary algorithm with reward-normalized stochastic selection (DMOEA-RNSS) is introduced to optimize performance within the defined operating window, generating dynamic Pareto fronts (PFs) and Pareto sets (PSs). Ultimately, the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) is applied to evaluate the PFs and determine the optimal operating strategy. The framework has been successfully applied to the optimal operational decision-making for a M701F gas turbine. The experimental results show that the Root Mean Squared Errors (RMSEs) between the predicted gas turbine performance state parameters and the actual operating measurement data are all less than 0.5 %. Furthermore, the variable load path from empirical gas turbine data remains within the established operating window's feasible boundaries, confirming window feasibility. Relative to the variable load path, the optimized strategy achieves a 0.1871 improvement in relative comprehensive energy efficiency (1.92 % in real), a 0.3652 reduction in relative emissions (28.34 g/s in real), and a 0.0674 reduction in relative economic indicators (0.7803$/s in real).