This thesis, based on a project realised in cooperation with Électricité de France (EDF), proposes a new concept for a Decision Aid Function FOr Restoration (DAFFOR) of transmission power systems after a blackout. DAFFOR is an interactive computer tool which provides the operators in power system control centres with guidance concerning the actions to execute during the restoration, in real-time conditions. In other words, it takes into account the real-time state of the power system, including the unforeseen events that may happen during the restoration. Since time is a limiting factor and the decision making is a highly combinatorial problem, a knowledge-based system is proposed in order to solve it. The restoration process can be decomposed into two main stages. The first one, skeleton creation, consists of starting the production units and connecting some transmission devices in order to energize a strong network. The second stage, load pickup, aims to supply the consumers. In DAFFOR, EDF's strategy for the first restoration stage has been implemented, and a new strategy for the load pickup stage has been proposed and implemented in the form of rules. The above restoration strategies represent DAFFOR's knowledge, which has been enhanced with a number of heuristics. DAFFOR consists of two kernels: the Reasoning kernel and the Real Time Update kernel. The Reasoning kernel has the task of assisting the operator during the restoration process and is the interactive guidance part of DAFFOR. It can either suggest a control action to execute on the power system to the operators or assess a control action provided by the operators. The control action is suggested with respect to operating limits (over- and under-voltages, frequency excursions and overloads) and according to knowledge (restoration strategy and heuristics). The feasibility of an action is tested within an internal dynamic simulator, which also takes into account the time necessary to physically execute an action (e.g., telephone a person in the field). The Reasoning kernel can adapt its operation via data generated by the Real Time Update (RTUpd) kernel. The RTUpd kernel steadily reads real-time power system data from System Control and Data Acquisition (SCADA) function and those entered by the operators (if unavailable from SCADA). It generates a coherent data set, which is the only real-time information available to the Reasoning kernel, and the message which indicates to the Reasoning kernel how to continue its operation. In addition to the real-time data, the RTUpd kernel has two feedback inputs internal to DAFFOR: a coherent data set generated in the previous data processing by the RTUpd kernel itself, and a simulated data set generated by the Reasoning kernel (i.e., its internal dynamic simulator). With these three inputs, the RTUpd kernel generates the current image of the power system, and identifies unforeseen events. Thanks to the RTUpd kernel, the Reasoning kernel may keep up with the dynamic evolution of the power system. The stand-alone prototype of DAFFOR has been tested with data provided by EDF, and shown very good efficiency. At present, it is about to be coupled with the EDF's operator training simulator in order to test its real-time functionality. This work also proposes an original method aimed at the determination of a strategy for the load pickup stage. A genetic algorithm has been developed which generates the optimized sequences of manoeuvres for different initial states of the power system for the second restoration stage. It uses the dynamic simulator as its evaluation function. The obtained results have shown that some additional manipulations should be done in order to deduce generic rules for the load pickup strategy. At present, the obtained sequences are classified in a decision tree, which permits the most adequate sequence for the initial state to be chosen.