A numerical method for constrained optimization
An iterative technique is developed to solve the problem of minimizing a scalar function f(x) subject to certain nonlinear constraints g(x) = O. The variables are separated into the basic variables x and the independent variables u. Each iteration consists of a gradient phase and a restoration phase. The gradient phase involves a movement (on a surface that is linear in the basic variables and nonlinear in the independent variables) from a feasible point to a varied point in a direction based on the reduced gradient. The restoration phase involves a movement (in a hyperplane parallel to x-space) from the nonfeasible varied point to a new feasible point.
The basic scheme is further modified to implement the method of conjugate gradients. The technique is generalized to optimization problems for dynamic systems, treating the equations of motion as the equality constraints.
Several problems in nonlinear programming and optimal control of dynamic systems are selected from the literature in order to demonstrate the practicability and validity of the method. In the case of parameter optimization, the work required in the restoration phase is considerably reduced when compared with the existing methods. When applied to the optimization of dynamic systems, this method requires only one integration of the equations of motion during the gradient phase. Moreover, arbitrary specification of the step length is not required.
Thesis76b.M536.pdf
1.75 MB
Adobe PDF
5b25fa65042e91679c1d0d74f0b03197