Strict environmental regulations demand gas turbine operation at very low equivalence ratios. Premixed gas turbine combustors, operated at very lean conditions, are prone to thermoacoustic instabilities. Thermoacoustic instabilities cause significant performance and reliability problems in gas turbine combustors, so their prevention is a general task. Splitting the fuel mass flow between different burner groups, i.e. using a burner group fuel staging technique, is a possibility to control the thermoacoustic instabilities. The resulting combustion perturbations have also effects on the gas turbine NOx emissions making it necessary to find an optimum balance between pulsations and emissions. This paper presents a model based active combustion control concept for the reduction of pulsations and emissions in lean premixed gas turbine combustors. The model is integrated in an observer structure derived from a Luenberger observer. The control logic is based on a PID algorithm allowing either the direct command of the pulsation level with a continuous monitoring and a potential limit setting of the NOx emission level or vice versa. The gas turbine pulsations and emissions are modelled using Gaussian Processes. - Gaussian Processes are stochastic processes related to Neural Networks that can approximate arbitrary functions. Based on measured gas turbine data they can be implemented in an easy and straightforward manner. The model provides the control system with real time data of the outputs resulting from settings of the staging ratio that is the actuating variable of the system. A model based control concept can significantly alleviate the effects of time delays in the system. The model based control concept allows for fast adaptation of the burner group staging ratio during static and transient operations to achieve an optimum of the pulsation and emission levels. During tests the model based control concept gave good results and proved to be robust even at high disturbance levels.
Gaussian Processes Model-Based Control of Underactuated Balance Robots
