In this paper, a multi-lane cellular automata model is adopted to investigate the evolution of traffic flow in a two-off-ramp system. The variable message sign (VMS) is used to realize traffic flow guidance. More attention is paid to the decision-making behaviors of the driver, i.e., choosing the target off-ramp and its influence on road capacity. The original model without VMS and the improved model with VMS are proposed respectively. Through numerical simulation, the corresponding phase diagram, the capacity, the flux of each section and typical patterns are obtained, then the effects of the injection probabilities, the fraction of leaving vehicles and the length of the exit lane on capacity are analyzed in detail. When the length of the exit lane is rather short but the entering probability and the fraction of leaving vehicles are rather large, the road capacity is increased noticeably with the help of VMS and therefore the state of traffic flow is improved. It is shown that the utilization efficiency of downstream off-ramp is enhanced and traffic jams are alleviated due to the guidance of VMS.