The gas-solid fluidized bed has wide application in industry such as refinery, metallurgy, and ore calcination. In the practical operation, it is usually combined with other units like conveying pipe, cyclone, downer and valves to run in a full-loop way. Compared to simulation of a single operating unit, the full-loop simulation of the gas-solid circulating fluidized bed system can reveal interactions between different operating components and diagnose the sudden change of operation, thus being of greater importance in industrial operation. In this study, the full-loop simulation and stability analysis of a gas-solid circulating fluidized bed of virtual process engineering (VPE) are carried out under the framework of two fluid model and kinetic theory of granular flow. The simulation shows that there exists periodic fluctuation of solids volume fraction and pressure drop in the riser where two distinct fluidization states, i.e., dilute fluidization and dense fluidization, appear alternatively, because of the occurrence of gas bypassing. To figure out the underlying cause, the influence of model factors (mainly refer to gas-solid drag) and operating parameters (i.e., solid inventory and superficial velocity) on the periodic fluctuation phenomenon is numerically investigated. It is found that changing the drag model cannot eliminate the fluctuation, while reducing the gas velocity and increasing the solid inventory are conducive to the stability of particle circulating transportation and avoiding the occurrence of "gas bypassing" because of the increase in pressure drop of the Loop-seal. On this basis, the Loop-seal valve which is reported to be closely related to particle conveying is focused on. The simulation with adding a virtual valve in the middle of the inclined pipe is performed and shows that the resistance of particle conveying increases thus ensuring the enough pressure drop of the Loop-seal to operate the full-loop system steadily. The time-averaged axial profile of pressure drop predicted by the simulation agrees with the experimental data. This method is found to be helpful to improve the stability of the full-loop simulation.