Event-Based Information Acquisition, Learning, and Control in High-Dimensional Cyber-Physical Systems

Abstract:

The objective of this project is to develop a theoretical framework for stochastic learning, decision-making, and control in high-dimensional cyber-physical systems. In our general framework, decision makers dynamically refine their estimates of the time-varying physical system based on acquired information, which may be obtained by distributed sensors. Such information acquisition enables a large scale/high dimensional yet scalable state estimation at the cyber system level which, in turn, is an enabler for the optimized utilization and control of the cyber-physical systems. Both centralized and distributed information acquisition and control will be considered, and the performance tradeoffs between these methods determined for various applications of interest. Applications for the proposed framework include smart buildings, intelligent transportation, energy-efficient data centers, and the future smart-grid

Our work over the past year has focused on developing this theoretical framework for a specific application: the joint path and charging decision problem faced by Electric Vehicle owerners. Specifically, we consider a system where EV drivers have heterogeneous travel demands, battery consumption characteristics, cost preferences, etc., and move accross the transportation network to fulfill their travel needs. The geographical flexibility of load that results from the mobile nature of EVs transforms the EV load control problem to a high-dimensional network- level problem that is not naturally decentralized anymore. This year, we first considered the decision problem of an individual EV owner who needs to pick a travel path including its charging locations and associated charge amount under time-varying traffic conditions as well as dynamic location-based electricity pricing. We show that the problem is equivalent to finding the shortest path on an extended transportation graph. In particular, we extend the original transportation graph through the use of virtual links with negative energy requirements to represent charging options available to the user. Using these extended transportation graphs, we then study the collective effects of a large number of EV owners solving the same type of path planning problem under the following control strategies: 1) a social planner decides the optimal route and charge strategy of all EVs; 2) users reach an equilibrium under locationally-variant electricity prices that are constant over time; 3) the transportation and power systems are separately controlled through marginal pricing strategies, not taking into account their mutual effect on one another. We numerically show that this disjoint type of control can lead to instabilities in the grid as well as inefficient system operation. Next steps in this work include reduced-dimension control, optimal clustering of demand and of travel paths for reduced complexity, stochastic modeling of prices, and profit-maximizing retail price design for aggregators. In subsequent years the PIs will investigate a broader class of complexity-performance tradeoffs in reduced- dimension information acquisition and state estimation. In addition, we will consider remote state estimation and control in Networked Control Systems with a constrained shared bandwidth, where we must optimized bandwith allocation to ensure the best state estimation for control under restricted communication. Finally, we will explore joint estimation and control over event-based information acquisition policies.