Inverse stochastic optimal controls (Q2681368): Difference between revisions

Consider a stochastic optimal control problem based on a diffusion-type controlled stochastic differential equation and an expected performance function defined by the sum of the costs \(f(t,X(t))\) along the trajectory, the terminal costs \(g(X(T))\) and a penalty term \(\theta\|u(t)\|^2\) from the control function \(u(t)\) with a certain weight parameter \(\theta\). Given an optimal solution \(u^*(t),X^*(t)\) of the stochastic optimal control problem and all information about the state equation and the objective function up to the unknown parameter \(\theta\), the problem is to determine the parameter \(\theta\). The inverse problem is reduced to a root finding problem for the unknown parameter, and a numerical method is given by approximating the expectation by the arithmetic mean of observed optimal controls and the corresponding optimal state functions. Numerical experiments are presented.