Due to the inherent tendency to be blocked by obstacles, reliability is a major challenge for visible light communication (VLC). The intelligent reflective surface (IRS) is an effective way to reduce or eliminate the influence of blockage in the VLC system. However, the complexity increases correspondingly due to complex shadow analysis when access points (APs), IRSs, and obstacles coexist. We proposed a robust layout optimization scheme for the IRS-VLC system to resist blockages. First, we model the random obstructions based on spatial geometry methods. Second, we build the optimization problem model considering consistent illumination, achievable data rate (ADR), the positions of the APs, and the directions of the IRS array. In order to achieve this objective, we develop an anti-occlusion method based on the non-dominated sorting genetic algorithm-II (NSGA-II) to get the Pareto front and use enhanced measurement functions (ASFs) to extract the best solution. Simulation results show that the proposed scheme works well in the IRS-VLC system. It is noteworthy that the rectangle layout has always demonstrated superior performance in the IRS-VLC system compared to other traditional layouts.