Comparative Study Between Three Modeling Approaches for a Gas Turbine Power Generation System

This paper presents a comparison between three modeling approaches for a gas turbine power generation system. These approaches involve artificial neural network (ANN), state space subspace and physical-based methods. The purpose is to compare the adopted methods in terms of modeling, simulations and control systems feasibility. It is proved that ANN is the most accurate methodology in reflecting constant outputs and large variation trends as the ANN is found to be able to capture the severe nonlinearity of the process easily. However, the state space is found be more feasible than other techniques for control system stability studies and applicability of control system algorithms in addition to best simulating small variation trends of the output, such as frequency excursions and temperature variations. The method used for state space system identification is based on the standard realization theory of controllability and observability matrices with the use of singular value decomposition technique to compute the system parameters. The superiority of the physical-based model is the acquisition of the physical insight necessary to study the system abnormalities, such as transient stability studies of the generator, and keeping better performance for simulating small trends or excursions even in the verification phase. The physical laws are rooted from thermodynamic relations, torque balance equation that governs the turbine-generator interactions, and the two-axis relations of the rotor and stator dynamics, and the physical model parameters were identified using genetic algorithm. The comparison that justifies the diversity in the capabilities of the models has been reported for guidance in future research.