To address the challenge of safe formation transformation of unmanned surface vehicles (USVs) in unknown complex environments, this paper proposes a method based on multi-layer grid sequence stitching and superposition. (1) Establish an appropriate initial grid considering the kinematic characteristics, spatial distribution, and quantity of the USVs. (2) Extract static obstacle features and predict the positions of dynamic obstacles to create a dynamic, real-time environmental grid. (3) Build a grid for spatial connectivity and future feasibility checks based on current and future formation layouts. (4) The D*lite algorithm is used to plan the local grid paths, and the priority negotiation strategy resolves conflicts and establishes a smooth path grid with spatiotemporal disjunction. Multiple rounds of simulation experiments with various cluster formations demonstrated that the proposed method can quickly and efficiently produce safe and reliable formation transformation paths. This method discretizes complex environmental information into simple digital data, facilitating environmental data sharing under limited communication conditions. By confining movements to absolutely safe grid points, it significantly enhances formation transformation safety. Converting the spatial constraints and path optimization of USV swarm into multi-layer grid layout stitching and superposition enables multi-layer distributed synchronous computation, providing potential for real-time dynamic handling of large-scale formation transformation tasks. This study offers valuable reference for research and applications in safe formation transformation of USV swarm in large-scale and complex environments.