Amid scarce land resources, urban demand for compact storage is rising. To address this demand, the Puzzle-based Storage (PBS) System has emerged as an innovative dense storage solution. In response to this demand, the Puzzle-based Storage (PBS) system has emerged as an innovative dense storage solution. By utilizing a grid structure where storage units move to adjacent escorts, the PBS system effectively enhances space utilization. However, existing retrieval algorithms for PBS systems are often complex and time-consuming. Furthermore, they overlook the impact of equipment directional switching on retrieval time, resulting in optimization approaches that are not fully aligned with real-world conditions. This study proposes a retrieval path optimization model for PBS systems of varying scales. An integer programming model is first constructed to incorporate directional switching, integrating directional constraints and state updates into the optimization process. To address the challenges of large-scale problems, a decomposition-based solving method is further developed. This method decomposes the original problem into subproblems and iteratively solves them using a strategic cycle. Experimental results demonstrate that incorporating directional switching significantly reduces system retrieval time. Moreover, the proposed model and method efficiently solve problems across different system scales, providing a novel approach to solving retrieval problems in PBS systems.