Demonstration of a Multi-Agent-Based Control System for Active Electric Power Distribution Grids Alexander Prostejovsky, Munir Merdan * and Georg Schitter Automation and Control Institute (ACIN) Vienna University of Technology Vienna, Austria Email: {prostejovsky,merdan,schitter}@acin.tuwien.ac.at Aris Dimeas Electric Energy Systems Laboratory National Technical University of Athens Athens, Greece Email: adimeas@power.ece.ntua.gr Abstract—The rapidly growing number of distributed energy resources and other kinds of active electric grid components, the limitations of the present electric grid infrastructure and the increased complexity of the networks that comes along with these challenges require new sophisticated control methods for future electric distribution grids. To cope with these challenges a control design is necessary that offers autonomous operation and scalability. This contribution shows the results of the first im- plementation of a Multi-Agent-System-based Smart Grid control system approach in a real laboratory environment. A so-called islanding case is considered where the laboratory grid equipment gets separated from the utility grid and reconnected again. The agents of the developed control system conduct their assigned equipment to react to the changed situation appropriately, hence demonstrating the control system’s applicability on a small-scale electric grid configuration. I. I NTRODUCTION In the presence of the continuous and ever-growing electric energy consumption, current electric grids with their dated infrastructure face many challenges, such as the limited ca- pacity of transportation and distribution, which leads to such situations where potentially available energy remains unused as it can neither be transported nor stored. Furthermore, the large- scale integration of distributed energy resources (DERs) in the low- and medium-voltage grids overturns the existent vertical production, transportation and distribution scheme of electric energy as stated by Ipakchi and Albuyeh [1]. Hidalgo et al. give examples for DERs, like photovoltaic (PV) systems, wind farms, small gas-turbines and biomass generators along with other active grid components such as intelligently controllable loads, novel storage systems [2], Smart Meters, etc. [3]. As the share in production of electric energy shifts from bulk generators to many small DERs, current centralized control systems along with their Supervisory Control and Data Acqui- sition (SCADA) facilities do not longer suffice to handle the increasing number of active grid components, and fundamental problems of centralized systems emerge according to Beidou et al. [4]. The large number and distributed character of future energy grid equipment requires control designs that are scalable and easily extendible in their functionality in order to support further developments. A distribution grid that incorpo- rates both novel active and intelligent grid components along a control system that meets the before mentioned requirements is also referred to as a Smart Grid (SG). A comprehensive * Present address: AIT Austrian Institute of Technology. survey along with a proposed roadmap for the development of such future grids is provided by the International Energy Agency [5]. Many features are attributed to SGs such as autonomous operation, optimization and reconfiguration al- gorithms for power loss minimization and voltage quality, demand-side management, etc., where several successful pilot projects are already being executed as illustrated by Brown and Suryanarayanan [6] and Stifter et al. [7]. However, as Solanki et al. point out [8], most of those novel control systems are still centralized and therefore they mark a single-point-of-failure which could potentially tie up the operation of the whole grid. Furthermore, widely used algorithms such as mathematical op- timization, optimal control and genetic and swarm intelligence algorithms as mentioned by Nelles [9] as well as Russel and Norvig [10] suffer from their limited scalability, because the calculation demand heavily increases with the optimization space, i.e. the number of electrical control variables. To address both of these issues, the power system control should be modularized and distributed over the grid, hence rendering it invariant to the scale and making it less vulnerable for information and communication infrastructure (ICT) single- point-of-failures. Multi-Agent-Systems (MAS) are an obvious and promising choice as the basis of a SG control system, since they overcome the threat of single-points-of-failures due to their distributed characteristic, and they have proven to be suitable for addressing the demands of SGs both theoretically by McArthur et al. [11], [12], and practically as shown by Chouhan et al. [13], Dimeas and Hatziargyriou [14], [15] as well as Nagata and Sasaki [16]. However, most solutions are still dependent on a central and ubiquitous management agent which itself depicts a weak spot. Therefore, other approaches are desired that take full advantage of the possibilities offered by MAS. The main objective of this work is to present the first implementation of such a novel SG control approach [17], [18] in a real laboratory environment in order to demonstrate its scalability, as no changes to the system architecture are necessary to adapt the control system to the laboratory equip- ment. The paper is organized as follows. Section II gives general information about MAS and the developed MAS-based control system architecture, while the agent design is covered in Section III. Section IV illustrates the actual test case along with the corresponding simulation model and the experimental setup. The results of the experiments are discussed in Section V, and conclusions are given in Section VI.