Simulation-Based Optimization as a Service for Dynamic Data-Driven Applications Systems

Abstract

Dynamic data-driven applications systems (DDDAS) must be adaptive in the face of highly fluctuating and uncertain environments. An important means to such adaptability is through the use of simulation models which can be leveraged for dynamic decision support. To provide high quality decision support, one can use simulations in an optimization loop to derive the best values of system parameters for a given system state particularly when the system has too many parameters and traditional means to optimize the outcomes are intractable. To that end, simulation-based optimization methods have emerged to enable optimization in the context of complex, black-box simulations thereby obviating the need for specific and accurate model information, such as gradient computation. An important challenge in using simulation-based optimization is optimizing the decision parameters. However, to ensure scalability and real-time decision support, one must be able to rapidly deploy simulation-based optimization in a way that makes the best use of available computing resources given the time and budget constraints. To address these needs, we propose a cloud-based framework for simulation-based optimization as a service (SBOaaS) to enable a flexible and highly parallelizable dynamic decision support for such environments. We illustrate the framework by using it to design a dynamic traffic light control system through simulation-based optimizations using the Simulation of Urban Mobility (SUMO) traffic simulation model that adjusts to the observed vehicle flow.