EVALUATION OF THE PARAMETERS USED IN ITERATIVE DYNAMIC-PROGRAMMING

To apply iterative dynamic programming (IDP) to optimal control proble ms having a very large number of control variables the use of randomly chosen values for control at each grid point is required. To gain ins ight into the effect of the number of allowable values for control, th e region contraction factor, and the number of grid points for the sta te vector to be used, computational results are presented for two nonl inear systems, one of which possesses numerous local optima. The relia bility of obtaining the global optimum for the bifunctional catalyst b lend optimization problem was found to be somewhat higher by using ran domly chosen values for control rather than by choosing the control va lues over a uniform distribution. The global optimum is obtained even when a small number of allowable values for control at each grid point and a small number of grid points for the states are used. There is a wide range of the region contraction factor for which rapid convergen ce to the optimum is obtained. Also the number of grid points for the state can be very small without adversely affecting convergence to the optimum.