Analysis of Various Partitioning Strategies for Multi- Agent System-Based Real-Time Load Management for NG IPS Ships Xianyong Feng, Karen L. Butler-Purry, and Takis Zourntos Dept. of Electrical and Computer Engineering Texas A&M University College Station, TX 77843, USA Email: fxy8410@neo.tamu.edu, klbutler@tamu.edu, and takis@neo.tamu.edu Abstract—The real-time load management techniques for next generation integrated power systems (NG IPS) for ships are being developed to balance the load demand and the power generation while satisfying the operating constraints of the system in real-time. To solve the real-time load management problem using a multi-agent system cooperative control protocol, the NG IPS must be partitioned into smaller subsystems that are modeled using dynamical agents. In this paper, three potential partitioning strategies are discussed along with their advantages and disadvantages. Unlike the first partitioning strategy modeling half a zone as an agent which aggregates a group of loads together, the last partitioning strategy models an agent for each electrical component and includes more system dynamics, which significantly increases the agent model accuracy. The results of the studies of the partitioning strategies can be utilized during the cooperative controller design to achieve real-time load management for NG IPS ships. Index Terms—Cooperative control, multi-agent system, next generation integrated power system, partitioning strategy, real- time load management I. INTRODUCTION The Power System Automation Lab at Texas A&M University has been conducting research on real-time load management techniques for NG IPS ships for several years [1], [2]. Real-time load management techniques can be used to balance the load demand and the power generation of power systems while satisfying the operating constraints of the system in real-time [1]. Since the NG IPS is a very complex dynamical system which is hundreds to thousands of orders, it is difficult to solve the real-time load management problem using a single controller or traditional optimization methods. A multi-agent system aims to cooperatively achieve the group objectives which are difficult to reach by a single agent or controller [3]. Thus, a multi-agent system cooperative control protocol is being developed to cooperatively achieve the real- time load management objectives based on a group of dynamical agents. In order to use the dynamical multi-agent system, the power system needs to be partitioned into smaller subsystems, which can be managed using dynamical autonomous agents. The nature of the system such as the topology of the system indicates a logical way to partition the system, which can make the multi-agent system more effective. The NG IPS system includes gas turbine generators, transformers, cables, power converters, propulsion loads, and DC zones which have similar topologies. If the notional NG IPS model is partitioned based on the topology of the NG IPS such as DC zones or DC-DC converters, a group of homogeneous subsystems is obtained. On the other hand, if the system is partitioned based on the individual electrical components, a group of heterogeneous subsystems is obtained. For the homogeneous multi-agent system, each agent has an identical dynamical model which may be modeled as a single integrator system [4], double integrator system [5], coupled- phase oscillator system [6], non-holonomic agent system [7], etc. The multi-agent system cooperative control protocol for the lower order (no more than second order) agent model has been systematically studied [3], [8]. For the higher order (third order or higher) agent model, only consensus algorithms [9], [10] and formation control [11] for linear agent model have been studied. For the heterogeneous multi-agent system, each agent has a different dynamical model which makes the cooperative control design more complicated. The formation control and cooperative control for the heterogeneous multi-agent systems have been studied in [12], [13]. However, these control methods can only be used to handle certain types of dynamical agent models. The notional NG IPS model includes different individual electrical components such as generators, propulsion loads, converters, and motor loads, which have totally different types of dynamical models. Thus, there are This work was supported in part by the Office of Naval Research under Grant N00014-09-1-0579. 978-1-4244-9273-2/11/$26.00 ©2011 IEEE 173