A multiple-model framework for parameter and state estimation of nonlinear systems
Michelle S. Chong
Time: Thu 2017-10-19 10.15
Location: Automatic Control Department
Abstract: Estimating the parameters and states of a system is crucial in any control strategy where only some states/parameters can be measured. Whilst popular stochastic estimation algorithms such as the Kalman filter and particle filters have been employed in this context, to the best of our knowledge, convergence is only guaranteed for local initialisations. In this talk, I will present a multiple-model framework for parameter and state estimation of nonlinear systems. The framework employs a bank of state-only observers, where each observer is designed for a sample of the known parameter set. Based on a selection criteria, one observer is chosen to provide the corresponding state and parameter estimate at each point in time. We provide convergence guarantees for the estimates for all initial conditions, provided that the sampling of the parameter set is adequate and a persistency of excitation condition is satisfied. I will also show how a class of Lipschitzian optimisation algorithms can be incorporated into the framework for improved performance. Lastly, I will touch on a motivation behind this work, which is to forecast the occurrence of seizures caused by Epilepsy and for close-loop stimulation for seizure abatement. The efficacy of the estimation framework is illustrated on a model of neuronal populations used to capture epileptic seizures.
Speaker bio: Michelle S. Chong is currently a postdoctoral research fellow in the Department of Automatic Control, Lund University. She completed a Doctor of Philosophy degree in the Department of Electrical and Electronic Engineering at the University of Melbourne. She has held postdoctoral positions in the same department, as well as in Prof. Joao Hespanhaʼs group at the University of California Santa Barbara. In 2013-14, she was awarded the American Australian Associationʼs ConocoPhillps postdoctoral fellowship to conduct research in UCSB. Michelle is also the joint-recipient of the best paper award in the 2016 ACM/IEEE 7th International Conference on Cyber-Physical Systems. Her main research interest lies in the design of estimation algorithms, with applications in secure control systems and in neuroscience.