NUMERICAL CONVERGENCE PROPERTIES OF ITERATIVE DYNAMIC-PROGRAMMING WHEN APPLIED TO HIGH-DIMENSIONAL SYSTEMS

Convergence of iterative dynamic programming (IDP), employing piecewis e linear control to establish the optimal control policy of very high dimensional smooth systems, was examined by considering two linear sys tems with quadratic performance indices. The convergence of IDP was fo und to be systematic and no problems were encountered in determining t he optimal control of a system having 250 state variables and 250 cont rol variables. Computationally, the use of IDP for the 250 dimensional system is as fast as establishing the optimal control policy by solvi ng the Riccati equation. For the second system consisting of a high-di mensional gas absorber, IDP was found to be faster than solving the Ri ccati equation when the number of plates was greater than twenty-five.