Optimizing Simulators: An Intelligent Analysis Tool for Complex Operational Problems
Abstract
The optimizing simulator represents a class of simulation tools in which the analyst can control the level of intelligence by adding information classes to the decision function. For example, the current MASS/AMOS simulator for airlift operations uses a simple rule-based function that acts purely on what is known at the time the decision is made, without using any forecasts of future activities. This is the first information class. The other three are: forecasts of exogenous events (classical forecasting), forecasts of the impact of a decision now on the future state of the system (for example, the impact of flying a C-17 into Saudi Arabia) and expert knowledge (although not reflect in the costs, an expert might tell you never to fly a C-17 into Saudi Arabia, or that it is best to use C5's when moving a certain type of cargo). Our approach to simulation bridges the traditional gab between simulation and optimization, and at the same time between operations research (which uses cost-based decision functions) and artificial intelligence (which uses rule-based decision functions). These techniques encompasses the current methods used in MASS (and its latest version AMOS), and at the same time can compete with commercial linear programming packages (which are used to solve models such as NRMO, which formulate the airlift problem as a linear program). We also allow the user to specify desired behaviors in the form of simple, low-dimensional patterns, which produces behaviors that may not be captured by a cost function. In this way, we provide a bridge between cost-based operations research models, and rule-based AT techniques.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 01, 2002
- Accession Number
- ADA405535
Entities
People
- Warren B. Powell
Organizations
- Princeton University