We consider direct numerical simulations of turbulent Rayleigh-Bénard convection inside two-dimensional square cells. For Rayleigh numbers Ra = 10^6 to Ra = 5 · 10^8 and Prandtl numbers Pr = 3 and Pr = 4.3, two types of flow regimes are observed intermittently: consecutive flow reversals (CR), and extended cessations (EC). For each regime, we combine proper orthogonal decomposition (POD) and statistical tools on long-term data to characterise the dynamics of large-scale structures. For the regime CR, centrosymmetric modes are dominant and display a well-defined dynamics, while the less energetic non-centrosymmetric modes (here, two double roll modes) fluctuate randomly. For the regime EC, a non-centrosymmetric mode is dominant and the overall dynamics display a stochastic behaviour. Furthermore, an additional simplification of the dynamics is proposed based on a K-means clustering of the POD phase-space. It reduces the problem to a network dynamics of finite number of clusters. This simplified analysis captures the main features of the CR and EC dynamics. Finally, we analyse within a range of Rayleigh numbers up to turbulent flow, the relation between dynamical regimes and the POD energetic contents as well as the residence time in each cluster.