To solve the problem of the increasing complexity and difficulty in effective management of the numerous equipment in the power dispatching data center computer rooms, an improved fast density peak clustering algorithm was proposed to perform deep analysis and mining of the massive operational data generated by the equipment in the computer room to accomplish the purpose of accurate fault location. Specifically, on the basis of the original fast density peak clustering algorithm(FDPC), an iterative overlay(OM) program with a decreasing local adaptive window(LAW) was used to build and establish a positioning map and a multi-dimensional density map. By combining the constructed density map with FDPC, the complexity of subsequent processing is significantly reduced. In addition, the proposed algorithm is insensitive to parameter changes, therefore, the selection process of cluster centers can be continuously optimized. The experimental results show that the proposed algorithm achieves a high accuracy rate for the fault detection of the equipment in the power dispatching data center computer room.