In this study, we performed a series of extremely long-duration (O(1e5) free-fall time units) numerical simulations of quasi-one-dimensional Rayleigh–Bénard convection with an aspect ratio of Γ = 0.1. The simulations spanned a narrow Rayleigh number range of 7.9e7 ≤ Ra ≤ 8.51e7 at a fixed Prandtl number of Pr = 4.34. Three distinct flow states were identified: time-invariant convection, quasi-periodic chaos, and aperiodic intermittency. By conducting ‘quasi-static' simulations with Ra(t) = Ra0 + ct, where c is the rate of change of Ra, we revealed three evident hysteresis loops that characterize the transitions between different flow states: a ‘normal' loop between convection and intermittency, an ‘inverse' loop between intermittency and chaos, and an ‘anomalous' loop between chaos and convection (with a long-lived metastable intermittent state). The existence of multiple hysteresis loops not only confirmed the memory effects in purely buoyancy-driven flows but also reshaped the understanding of non-equilibrium dynamical systems.