A framework for designing policies for networked systems with uncertainty

This paper presents a framework to design policies for networked systems. The framework integrates model building, stability analysis of dynamic systems, surrogate model generation and optimization under uncertainty. We illustrate the framework using a transportation network benchmark problem. We consider bounded rational users and model the network using software agents. We use Largest Lyapunov exponents to characterize stability and use Gaussian process model as an inexpensive surrogate, facilitating computational efficiency in policy optimization under uncertainty. We demonstrate scalability by solving a traffic grid policy design problem and show how the framework lends itself towards carrying out stability versus performance tradeoffs.