Optimizing Energy Consumption: from Wireless Sensor Networks to Large “Smart” Buildings
Committee Members: Advisor: Ioannis Paschalidis, SE/ECE; Appointed Chair: John Baillieul, SE/ME; Christos Cassandras, SE/ECE; David Castanon, SE/ECE; Michael Caramanis, SE/ME
Abstract: This dissertation addresses two problems of optimizing energy consumption in Wireless Sensor Networks (WSNETs) and “smart” buildings.
For the first topic, we study the energy-efficient implementation of averaging/consensus algorithms in WSNETs. For static topologies we notice that a Bidirectional Spanning Tree (BST) is preferable in terms of convergence time. We formulate the combinatorial optimization problem of selecting such a minimal energy tree as a mixed integer linear programming problem. We then devise a semi-definite relaxation and a series of graph based algorithms that yield energy-efficient BSTs, and establish associated bounds on the optimal cost. For dynamic topologies we consider a load-balancing algorithm which has preferable convergence time. We formulate the problem of selecting a minimal energy interconnected network over which we can run the algorithm as a Dynamic Programming (DP) problem. We first consider the scenario of a large enough time horizon and show that the problem is equivalent to constructing a Minimum Spanning Tree. We then consider the scenario of a limited time horizon and employ a “rollout” heuristic that generates near-optimal solutions.
For the second topic, we develop a market-based mechanism that enables a building Smart Microgrid Operator (SMO) to offer regulation service reserves and meet the associated obligation of fast response to commands issued by the wholesale market Independent System Operator (ISO). The proposed market-based mechanism allows the SMO to control the behavior of internal loads through price signals and to provide feedback to the ISO. A regulation service reserve quantity is transacted between the SMO and the ISO for a relatively long period of time. During this period the ISO follows shorter time scale stochastic dynamics to repeatedly request from the SMO to decrease/increase its consumption. We model the operational task of selecting an optimal short time scale dynamic pricing policy as a DP that maximizes average SMO and ISO utility. We then formulate a non-linear programming static problem that provides an upper bound on the optimal utility. We study an asymptotic regime in which this upper bound is shown to be tight and the static policy provides an efficient approximation of the dynamic pricing policy.