Systemizing the Solution of Simulation-Driven Optimization Problems

This talk introduces a generic computational framework to aid the numerical solution of optimization problems constrained by differential equations, so-called "Simulation-Driven Optimization Problems" (SDOs). Though SDOs can be prescribed in a variety of ways, their numerical solution often involves the same set of operations. This insight led to the creation of "TSOpt", the "Time Stepping for Optimization" software package, which systemizes the solution of SDOs (via first-order optimization methods) using a base simulator. TSOpt encapsulates time-stepping simulations and orchestrates their execution, forming data structures needed by the optimization algorithm (e.g., the gradient). Further, TSOpt implements consistency tests, which ensure that the proper relationships between the various simulators hold. I will conclude the talk by using TSOpt to solve the optimal well rate allocation problem, which is an optimization problem with oil-reservoir simulation constraints.

This is joint work with Bill Symes.

Presentation Slides: PDF

Note: Visitors from outside NIST must contact Cathy Graham; (301) 975-3800; at least 24 hours in advance.