The project FLOCCON is dedicated to the development and validation of an innovative strategy to accelerate solvers for fluid mechanics. The project focuses on incompressible solvers, containing two parts: (1) a linear Poisson equation, and (2) a non-linear advection equation. The key idea of this project is to use deep learning to train neural networks based on solutions of these two equations. To go further, the project will examine learning methods which can guarantee a target accuracy. To do so, physical-based and long-term 'loss functions' will be introduced, in order to ensure a limited error accumulation in time. Moreover, an hybrid strategy will be proposed to obtain a robust solver. finally, an optimisation of this new network-based solver will be carried out on CPUs/GPUs. In addition to classical validation cases, a target application will be simulated on the pollutant dispersion in a large city, which is a challenging case for classical HPC solvers.