Artificial Intelligence
Natural language processing
Semantic web

Decision-support support for electrical distribution network

Student: Mohamed Gaha

Supervisor: Michel Gagnon

Co-supervisor(s): Frédéric Sirois

Nowadays, the information system technologies are increasingly used in power distribution systems to improve network reliability and performance. The impact of these structural changes is important and requires in-depth studies and investigations. A better understanding of the effect of these technologies is required to optimize the network. However, the simulation of power network is a complex task, where several technical issues need to be considered such as : electrical, mechanical, economical, natural and human aspects. The idea is to develop a multi-agent system (MAS) that can process complex simulations. Such a system is extensible and modular and it is composed by numerous simple agents that can collaborate and interact in order to achieve complex objectives. Multi-agent systems are capable of reaching goals that are difficult to achieve by monolithic systems or individual agents, which can be complex and hard to maintain and extend. Nevertheless, the development and the maintenance of a MAS is a complex task that has to be performed by experts on computer science and multi-agent systems. In the framework of the project LEOPAR, carried out by the extit{Institute de Recherche d'Hydro-Québec}, we have as a main objective to develop an accessible and comprehensive MAS. The project's aim was to allow managers to modify the behavior and the objectives of the simulator without the assistance of an expert. To this end, we developed a simulator based on Blackboard and MAS. Our system relies on a common pool of data to share information between agents. This type of mechanism reduces the communication complexity and makes the development of agents easier. In addition, we defined a new action language that allows to incrementally describe the agent's actions, effects, conditions and relations. Our action language is automatically translated into a non-monotonic logic (Answer Set Programming) in order to process the agent's actions. The translated answer set program has shown to be effective in providing action plans. The action language combined to answer set programming allowed us to develop a powerful and accessible simulator, enabling novice to add, change, and remove agents' behavior. Our simulator works properly and allows, among other things, processing power network assessments using a Monte-Carlo approach. It analyses the impact of introducing new types of technologies, by comparing performance indicators of the network. Moreover, it is able to simulate with accuracy a wide variety of phenomena as wire overloading, protection mechanism activation, tap changer changes, human intervening team patrols, restoration process and network reconfiguration. It has been tested on realistic distribution network of Hydro-Quebec and it performed well in assessing networks. Our simulator is performing similarly to a classical multi-agents system, but with the benefit of being accessible and easy to use.

For more information, click here



Answer Set Programming and Blackboard System

Mohamed Gaha, Michel Gagnon, Frédéric Sirois

International Workshop on Learning, Agents and Formal Languages, Lyon