Simulation-Driven Optimization (SDO) problems (also referred to as "optimal control" or "4D optimization") are optimization problems with a simulation constraint. In the past decade, SDO techniques have been established as a promising tool for aviation analysis. Given a parameter-dependent simulation model, SDO techniques can automatically determine the optimal parameters that yield the desired simulation behavior. SDO techniques have been applied to aviation problems such as flight trajectory optimization, air traffic flow design and safety analysis of auto-land systems. The algorithmic solution of an SDO problem requires communication between simulation code (e.g., the numerical solution of the equations of motion) and optimization code (e.g., the Newton method). Typically, multiple simulations must be performed to form the numerical derivative of the cost function we seek to minimize or maximize, which must then be passed to some optimization software. This paper introduces the "Time-Stepping for Optimization" software framework or TSOpt to aid solution of SDO problems. TSOpt orchestrates communication and data exchange between the optimization code and the simulation code. TSOpt also offers support for implementation variants of the adjoint state method, a numerically efficient way to form derivatives for SDO problems. Further, TSOpt is equipped with tests that help ensure the correct numerical solution of SDO problems. Besides a concrete C++ software package, TSOpt framework offers a software paradigm that can be used to solve SDO problems on any platform, and in any language. I demonstrate this claim by solving a exploring the effect of low-fidelity wind data for a trajectory-based optimization problem in MATLAB.
Mendeley saves you time finding and organizing research
Choose a citation style from the tabs below