REDUCED HESSIAN SUCCESSIVE QUADRATIC PROGRAMMING FOR REALTIME OPTIMIZATION

Abstract

Reduced Hessian Successive Quadratic Programming (SQP) is well suited for the solution of large-scale process optimization problems with many variables and constraints but few degrees of freedom. The reduced space method involves four major steps: an initial preprocessing phase followed by an iterative procedure which requires the solution of a set of nonlinear equations, a QP subproblem and a line search. The overall performance of the algorithm depends directly on the robustness and computational efficiency of the techniques used to handle each of these sub-tasks. Here, we discuss improvements to all of these steps in order to specialize this approach to real-time optimization. A numerical comparison of reduced Hessian SQP with MINOS (Murtagh and Saunders, 1982, 1987) is provided for the optimization of the Sunoco Hydrocracker Fractionation Plant (Bailey et al., 1992). The case study consists of about 3000 variables and constraints and includes several scenarios related to parameter estimation and on-line process-wide optimization. A study of the effect of optimizing the DIB distillation column which constitutes a subproblem of the Sunoco example is also included. The results indicate that our algorithm is at least as robust and an order of magnitude faster than MINOS for this set of problems.