Monitoring of combustion oscillation by cascade extended state observer

In modern gas turbines, active control has proved to be a useful tool in the suppression of combustion oscillation. As the basis of early warning and control, signal monitoring will directly affect the performance of the whole system. Traditionally, time series analysis and machine learning methods are often applied to monitor system information. However, there are some drawbacks, such as a large amount of calculation, limited accuracy and insufficient generality, which limit the engineering application. To solve these problems, this paper presents a monitoring method based on extended state observer (ESO), which utilizes the observed derivative information for online monitoring. Due to the difficulty in parameter tuning and sensitivity to noise, conventional high order ESO is modified to cascade low order ESO. Meanwhile, a compensation scheme based on Taylor expansion is proposed to tackle the phase delay caused by the cascade group. Combined with known system information, the structure of ESO can be further updated, which improves monitoring accuracy. Simulation results show that the proposed method can achieve computation simplicity, high monitoring accuracy, and has great robustness and noise immunity.