Fluid flows in nuclear reactor cores are highly turbulent and their numerical simulation is particularly demanding in terms of computing resources. Computational codes are therefore based on massive parallelism, which can take several forms. The first level, used in the TrioCFD code developed by CEA's Energy Division, consists of accelerating the resolution of the linear systems resulting from the discretisation of the equations, time step by time step. The space-time domain decomposition (DD) constitutes a second level of parallelism, the study of which is proposed in this thesis. By dividing the computational domain into sub-domains, the principle is to transform the physical problem into a set of decoupled sub-problems of smaller sizes, which can be solved in parallel over the entire time interval. Space-time data is then exchanged on the interfaces between sub-domains and the process is repeated until convergence towards the solution of the global problem. Finally, a third level consists of coupling the DD method with a time-parallel algorithm, which divides the time interval into temporal sub-windows and uses a prediction/correction strategy, with the costly correction phase being performed in parallel on each temporal sub-window. The proposed thesis, which will mainly be pursued at the CEA Saclay, has a theoretical aspect on the analysis of these methods when applied to the Navier-Stokes équations and a practical aspect for an efficient implementation in the TrioCFD code, by optimising the algorithmic of these different steps and their coupling.