Energy management is, nowadays, a subject of uttermost importance. Indeed, we are facing growing concerns such as petroleum reserve depletion, earth global warming or power quality (e.g. avoiding blackouts during peak times). Smart grids is an attempt to solve such problems, by adding to power grids bidirectional communications and ICT capabilities in order to provide an intelligent autonomic management for the grid.This thesis focuses on the management of microgrids thanks to multiagent systems (MAS). Microgrids are low-power networks, composed of small and decentralized energy producers (possibly renewable) and consumers. These networks can be connected to the main grid or islanded, this make them more complex. Due to their complexity and their geographical distribution, smart grids and microgrids can not be easily managed by a centralized system. Distributed artificial intelligences especially MAS appear to be a solution to resolve problems related to smart grids.Firstly we defined an approach implementing feedback loops. These feedback loops exist in complex systems which can be defined with several abstraction levels. Two levels are interacting. The micro-level contains a set of agents owning behaviours that can be combined. The result of the combination imapct the state of the system. The macro-level processes these influences to define a new state of the system which will imapct the agents behaviours at the micro-level.This feedback loop separates behaviours on several levels.This approach is used to defined a demand and supply matching problem in microgrid. This problem afford to manage a set of goals which currently are independently processed. Finally, an application is developed using MAS that ensures grid stability thanks to storage systems. This application was thought to be integrated to the approach detailed above.Secondly, a grid simulator id developed. This simulator allows dynamic control of devices. It is based on three main principles.• An agent-based model of the simulator to represent the complexity of electrical networks. • The use of the holonic paradigm to take into account the multiple levels inherent to power grids. • The use of the influence/reaction model and offers a technology that can manage simultaneous actions, possibly conflicting, in MAS.
