Spatial information sampling algorithm (SIS) is an evolutionary algorithm that uses intelligent solutions for decision making. In SIS, the shape of solution space of the population is normalized to a hypercube distribution. The SIS always searches around the best individuals, and thus suffers from a local optimum trapping problem. Therefore, we propose an improved spatial information sampling algorithm (ISIS). We innovatively propose a dynamic adjustment of the population structure strategy and improve on the original search strategy so that the ISIS algorithm can search around multiple good individuals simultaneously. We compare the results of ISIS on 29 benchmark problems from CEC2017 with other state-of-the-art algorithms, and our algorithm has significant advantages.