Although they are very promising, MAS are still an emerging technology in the
field of power systems. Topics to explore in the future include:
• Making agents smarter. Most agents used in MAS for power systems tend to be closer to reactive agents than to cognitive agents. Making them smarter would enable a greater autonomy and more complex decision-making processes with additional parameters taken into account. The learning capability of agents could for example have many applications, from forecasting to scheduling, and could partially pre-solve some problems. Similarly, the planning capacity of agent is rarely used.
• Fully distributing and automating the decision-making process. Decisionmaking processes are usually partially centralized, i. e., not fully distributed. If one or more central controller fails, or if a communication channel is cut [82], then the whole system operation may be compromised. Redundancy can partially help avoid such situations, but increases costs. Achieving a fully distributed decision-making process would solve this problem: if each agent takes his own decisions, sometimes after discussing with others, it would not be greatly impacted by the failure of another agent. However, distributing decision-making often comes at the cost of having more intensive communication.
• Creating a modeling and simulation tool to facilitate the development and testing of MAS for power systems. Current solutions require co-simulation between a MAS and a simulation tool such as Matlab/Simulink or PowerWorld Simulator [83]. A tool mixing both would highly simplify simulation, development and testing MAS for these applications.
• Testing MAS on very large-scale grids. In most research papers, MAS are tested on small-scale systems and show good results. However, very few work has focused on the scalability of MAS for such applications: how would the system perform when the grid size increases? Will communication and computation requirements explode? How will the system behave in case of perturbations or attacks? Such studies will be essential to determine the optimal decentralization degree to use, as well as the corresponding infrastructure costs for developing smart grid communication infrastructures.
• Testing MAS for control on real-scale industrial systems. So far, MAS have been mostly a subject of research in the field of power systems, but their industrial and commercial applications are rare, with exceptions such as the application presented in Sect. 15.5.4. Feedback and lessons learned from early experiments will be essential to foster the adoption of MAS technology in the industry. Three main difficulties have to be faced: utilities are sometime reluctant to investigate the use of MAS, which are still widely unknown due to a lack of practical experience; communication costs are high because MAS often require new equipment, especially at the distribution level; and power system standards are not written with MAS in mind. These difficulties also apply to most new smart grid technologies.
• Integrating user behavior models. As consumers are and will remain central players in smart grids, arises the need to model consumer behavior. In the future, so-called ‘‘prosumers’’ may take decisions using their home energy management interface and will directly impact how and when electricity is consumed, for example through demand response mechanisms that may affect consumers’ comfort level and billing, but also utilities. Qualitative studies provide important and useful information, but quantitative results are required to make regulatory and investment decisions, and to develop potentially successful new products. In parallel to modeling such consumer agents, real experiments will also have to take place to make these models more accurate and validate them.
• Securing MAS communication. A prerequisite to any commercial application is that systems must be secure and resist to attacks of all sorts. Although this problem is much wider than only for MAS, its importance was stressed by the recent concerns raised by worms such as Stuxnet, which aimed at disrupting industrial processes controlled by PLCs (programmable logic controllers). MAS thus have to include the most efficient and secure technologies before being deployed in the industry.
[1] 0.68
О
[3] 0.66
064
29.8 30 30.2 30.4 30.6 30.8
Time (s)
[4]dc2 Cdc2
[5] http://www. powermatchingcity. nl. Last visited June 27, 2011.
