B.S., University of Science and Technology of China, Automation,

M.E., University of Kentucky, Mechanical Engineering

Ph.D Candidate,  Purdue University, Aeronautics and Astronautics Engineering

 

Research area – System Identificaiton, Computational algorithms for Optimal control

 

1. Hybrid system: Statistical method for the identification of hybrid systems

Hybrid systems are dynamical systems with interaction continuous dynamics (modeled, for example, by differential equations) and discrete event dynamics (modeled, for example, by automata). A statistical method of inference and learning of a hybrid system given a sequence of observation data is to treat all unknown quantities as random variables, assign priors to these quantities, and then infers posterior probabilities given observed data. Methods for linear hybrid system identification include dynamical programming, neural network, variational learning, and MCMC. Currently, we are investigating the last two methods and propose algorithms to recursively identify jumped linear hybrid system parameters by using the output data.

Currently, I am working on the identification of hybrid systems, i.e. given a sequence of   observations and/or inputs, how to infer the parameters of the hybrid model. Traditional methods such as Kalman smoothing and Baum-Welch algorithm have been successfully applied to identify the parameters of continuous systems and discrete systems respectively.  However, to overcome the potential intractable problem due to the inter-correlations between the continuous dynamics and discrete dynamics in the hybrid system, recent researches have been focusing on to apply methods from other areas, such as dynamical programming, Bayesian statistics, and geometry algorithm etc., for the hybrid system identification. My interest is just to explore those methods to develop a new and effective identification algorithm, which I have been successfully applied the Variational approximation methods to identify a class of stochastic linear hybrid systems.

 

Contacts:

Room 337B

Potter building, Purdue University

West Lafayette, 47906

 

Email: li31@purdue.edu

Office phone: +1 765-496-6633