Variance-reduced model predictive control of Markov jump processes
P. A. Maginnis, M. West, and G. E. Dullerud
in American Control Conference (ACC), 2016.
We present an algorithm for variance-reduced Monte Carlo estimates of the expected cost-to-go used in the stochastic model predictive control of Markov jump processes. Specifically, we extend previous work on antithetic stochastic simulation of Markov chains with a finite number of reaction classes to the approximate computation of an expected cost function of a controlled process. In the presence of strict constraints on number of available Monte Carlo samples, we demonstrate significant reduction in the number of Monte Carlo simulations required to achieve a particular cost, including a factor of two reduction in the small resource limit, for a simplified, nonlinear chemical reaction model.
Full text: MaWeDu2016b.pdf