Jump linear quadratic control with random state discontinuities (Q1093612)

The author considers the linear stochastic system \(\dot x=A(t,r)x+B(t,r)u\) where x is the plant state vector in \(R^ n\), u(t) is the control vector in \(R^ m\), A, B are random matrices depending on the plant mode r(t). The evolution of r is representated as a Markov chain on a finite valuation set S. The performance index is defined as \[ J=E\{\prod^{t_ f}_{t_ 0}x^ T(t)Qx(t)+u^ T(t)Ru(t)dt| \quad t_ 0,x(t_ 0),r(t_ 0)\}. \] The matrices Q and R, being non- negative and positive definite, respectively, are also mode dependent \((Q=Q(r)\), \(R=R(r))\). For this problem the optimal regulator is obtained. It involves a modified set of coupled Riccati equations. The definitions and criteria of stochastic stabilizability and detectability are given to characterize the behaviour of the optimal system on long time intervals. An example is given to illustrate the obtained results.