In recent years, a number of major earthquakes have occurred around the world, which seriously threaten human life. The RoboCup Rescue Simulator System, a multi-agent system, simulates disaster scenarios to conduct collaborative rescue operations. Police agents play a crucial role in rescue operations, and the task allocation of police agents is one of the most important parts in obstacle clearing. However, existing clustering allocation methods for police tasks still require further efforts to enhance efficiency. To address this, this paper optimizes the clustering allocation strategy for police agents by combining the K-means + + algorithm for building clustering with the Hungarian algorithm for optimal task allocation of unburied police agents, enabling them to operate in different areas. Experimental results demonstrate that this strategy improves the rationality of police allocation and significantly enhances rescue efficiency across various map scenarios, offering a valuable reference for multi-agent collaboration in disaster rescue.