Real-Time Modelling of the Thermoacoustic Dynamics of a Gas Turbine Using a Gaussian Process

Abstract

Premixed gas turbine combustors operated at very lean conditions are prone to thermoacoustic instabilities. Thermoacoustic instabilities have negative effects on the operability of the combustion chamber. The prevention of thermoacoustic instabilities is a major design goal of the gas turbine combustor system as well as its control system. An appropriate real-time model helps the design of effective control algorithms for the prevention of thermoacoustic instabilities. This paper presents a black-box real-time modelling approach for thermoacoustic instabilities simulation using a Gaussian-Process. A Gaussian Process is a stochastic process that can approximate arbitrary functions, similar to Neural Networks, but with the advantage that it can be implemented and tuned in a more straightforward manner since a theoretical framework exists for the optimization of the hyperparameters influencing the process. The Gaussian Process can be trained in a fast and straight-forward manner. The trained Gaussian Process has been proven to be very efficient numerically, which enables it to be used in a real-time simulation environment. The real-time gas turbine model is to be used in the development of control algorithms that allow for low-NOx and robust operation of the gas turbine in conjunction with low acoustic pulsation levels. Verification on a gas turbine demonstrated the high accuracy of this modeling approach for a wide range of operating conditions. Moreover, it was shown that a Gaussian Process trained with data of one engine correctly reproduced acoustic pulsation behaviour of another engine.