Home
The last decade has seen dramatic advances in the development and deployment of technologies for sensors, mobile computing, and wireless communication networks. At the same time, use of on-line simulation as a means to better manage and control operational systems has been receiving increased attention. On-line simulations, fed by real-time data, can rapidly predict future system states for use in planning and system management. Applications arise in many diverse areas such as manufacturing, supply chain optimization, and transportation system operations, to mention a few. As mobile and ubiquitous computing and communications see increased deployment, embedded on-line simulation will be of growing importance in the years ahead, and potentially could become pervasive in everyday life.
We envision a world where embedded on-line simulations are ubiquitous and play a fundamental role in creating large, complex self-managed distributed systems. We refer to collections of autonomous, embedded on-line simulations that have overlapping interests in portions of an operational system as ad hoc distributed simulations. Ad hoc distributed simulations represent a fundamental shift in methods for modeling and predicting future states of operational systems. We explore the question: “can ad hoc distributed simulations incorporating real-time information be effective in rapidly producing reliable forecasts of future system states for use in system management and optimization?” As discussed later, ad hoc distributed simulations offer significant technical and economic advantages over conventional approaches such as global, centralized simulations. In the context of transportation systems, we envision an autonomously reconfigurable, continuously operating, dynamically extensible, real-time, embedded simulation environment that is distributed across in-vehicle computer systems, roadside computers, and traffic management centers. Autonomous predictive models such as in-vehicle agent-based simulations join the system as their vehicle begins a trip, obtain information from sensor networks and information systems, perform predictive functions on behalf of their clients, share their projections with other simulations as needed, and exit the system when their vehicle reaches its destination.
To illustrate the role of ad hoc distributed simulations in self-managed systems, consider the control of a dynamic, responsive intelligent transportation system (ITS). At the center of an ITS deployment is the ability to accurately and reliably capture the current system state and to predict future conditions; this is represented by the “modeling and prediction” component shown in Figure 1. Current state models provide a static picture of existing conditions based on real-time sensor data from in-field devices at multiple time resolutions (minute, hour, day, and season). Sensors such as inductive loops embedded in the roadway, video detection systems, and radar provide current and historic infrastructure performance data to the modeling and prediction subsystem. Onboard systems in instrumented vehicles can deliver current vehicle speed, location, and travel path information, as well as historical travel data. Enhanced models forecast future states, producing dynamic predictions via analytical approaches, simulation, or some combination. These forecasts are fed to system control and management decision makers; these may incorporate both automated procedures and real-time human interaction. System control and management decisions are then returned to the prediction process and update predictions to account for control adjustments, creating a cyclic process of prediction and control updates within some set of convergence rules. Also included in such systems is the potential that drivers who are informed of the predicted system states may change their travel behavior in response to this new information.
As an example, consider an emergency response scenario, such as a major chemical spill. Prior to the accident, distributed simulations onboard vehicles operate within the transportation infrastructure determining desirable routes for individual drivers, while simulations in traffic signal controllers and traffic management centers optimize signal timings. The desired response to a major chemical accident that results in roadway closures is to reroute traffic away from the spill, while accommodating inbound movement of emergency vehicles to contain the incident. Distributed simulations coupled with real-time sensor data will identify the lane closure problem through the observed rapid change in traffic flows and speeds and will reconfigure autonomous traffic control systems in response to the unforeseen events.
Figure 1. Overview of Intelligent Transportation System Control and Management
To study such an autonomously reconfigurable engineered system, our research integrates expertise from multiple disciplines, drawing from computer science and simulation, transportation systems analysis and modeling, traffic control and engineering, in-field data collection, statistical methods, driver behavior models, vehicle instrumentation, and wireless communication.