An Automated Process to Create Start Values for Gas Turbine Performance Simulations Using Neural Networks and Evolutionary Algorithms

Abstract

This paper presents a fully automated methodology to create start values for gas turbine performance computer programs. The methodology employs the application of evolutionary algorithms for a more robust convergence of the iterative process of a performance program as well as neural networks for a self-learning start value generation procedure. The achieved results showed that a connection of both methods for the creation of performance model start values is a feasible option. Different types of neural networks as well as several training methods have been evaluated in order to find the best equivalent model. Having found a practical approach for the structure of the neural networks, several performance models of different levels of complexity have been tested. The combination of both of the above presented steps achieved very good convergence rates in combination with a minimum effort for the creation of start values.