Sensitivity analysis of stochastic dynamical systems

Abstract

A new, efficient algorithm is developed for the sensitivity analysis of a general class of stochastic dynamical systems. The algorithm is based on an idea of the likelihood ratio method that utilizes a probability density information for the sensitivity analysis, and on the Fokker-Planck or Kolmogorov's forward equation for computing the evolution of probability densities. The ideas of stochastic sensitivity analysis and likelihood ratio method are presented and combined to derive the sensitivity of average values of the performance functional with respect to system parameters. The present algorithm avoids the time consuming Monte Carlo or stochasic simulations, and, instead, sensitivity gradients of the probability density function and performance functional are directly computed during single-run simulation. © 1992 Taylor & Francis Group, LLC.