The first part of this work proposes the use of an evolutionary algorithm in a dynamic context to resolve airspace conflicts in a control sector. In a first step, operators of the evolutionary algorithm are modified. In order to propose a continuous solution, a naive approach and an approach using a memory preserving the previously computed solutions of the algorithm are presented. Results obtained from the simulations highlight the advantages of using explicit memory and new operators to speed up successive solutions without affecting the quality and diversity of the answers provided. Furthermore, and despite the reuse of the same population of individuals, this approach proves to be robust when unexpected new constraints happen in the sector, which could simulate the decisions of air traffic controllers. The second part consists of learning and interpreting the uncertainty parameters of air traffic controllers in their aircraft conflict resolution task. Learning these uncertainties can help to calibrate the decision support tool as closely as possible to the operating modes of human controllers. This step is essential to make the tool acceptable and useful. The learning method introduced using an evolutionary algorithm was first validated on nominal data and then tested on experimental data. The integration of the parameters found into a conflict resolution algorithm led to solutions that are more similar to human solutions. Finally, we apply the method to real traffic data recorded at the Bordeaux Control Centre. This last step provided orders of magnitude for the uncertainty parameters.