报告时间：2019年12月22日（周日）16:00

报告地点：东校区信息馆107

报告题目：《On two non-asymptotic estimation methods》

报告人：Dayan L

报告人简介：After being postdoctoral fellow in Arts et Métiers ParisTech and at King Abdullah University of Science and Technology (KAUST) in Saudi Arabia, he becomes a tenured Associate Professor in INSA (French National Institute of Applied Sciences) Centre Val de Loire in 2013, where he belongs to Control Team in PRISME Laboratory. Dr. Liu’s main research interests concern with estimation and identification for integer order and fractional order systems. Until now, he has published about 60 papers in international journals and conferences such as IEEE Transactions on Automatic Control, Automatica, SIAM Journal on Scientific Computing and Systems & Control Letters, etc. Thanks to his Ph.D. work, he earned the Chinese Government award for outstanding selffinanced students abroad in 2012. Since October 2017, he is member of IFAC's Technical Committee: 2.2 Linear Control Systems. Since January 2019, he is member of CAA Technical Committee: Fractional Order Systems and Control. Since May 2019, he becomes editorial board member of the journal of Fractal and Fractional.

报告内容简介：

For cost and technological reasons, there always exist some variables and parameters which cannot be measured. Moreover, the measurements usually contain noises. Sometime, fast estimations with convergence in finitetime are required in online applications. For these reasons, the modulating functions method introduced by Shinbrot in 1954 and the algebraic parametric estimation method introduced by Fliess and SiraRamirez in 2003 both originally for system identification have been applied and extended in signal processing and automatic control, such as parameter estimation and numerical differentiation, etc. The two methods have the following advantages. Firstly, the obtained estimators are exactly given by integral formulae of the observation signal. Thus, they are algebraic and nonasymptotic. Fast estimation can be provided using sliding integration window with finite length. The knowledge of initial conditions is not needed and the derivatives of noisy signals don't need to be calculated. Moreover, thanks to the integrals in the formulae, they are robust with respect to corrupting noises without the need of knowing in priori their statistical properties. Recently, these methods have been extended to fractional order systems. In this talk, the ideas of these two methods will be explained by giving simples examples. Moreover, the applications to parameter estimation and numerical differentiation will be presented.