Abstract: The energy consumption of industrial parks is the key to the implementation of the national energy "dual carbon" strategy. Its carbon emissions and energy efficiency management are necessary to respond to the "dual carbon" goals and the dual control of energy consumption. In order to achieve carbon flow analysis and panoramic perception of carbon emissions in industrial parks, a detailed and accurate topological relationship model needs to be constructed. Therefore, this article proposes a high reliability topology recognition method based on load identification and feature propagation. Firstly, in the "device branch" layer, the KNN algorithm is used to construct an industrial equipment feature library, and equipment type identification and branch connection are achieved through electrical transient characteristics; secondly, in the "branch user transformation" layer, based on Kirchhoff's law and the longitudinal conduction mechanism of load characteristics, a threshold for determining the subordinate relationship between branches and transformers is established through the matching probability analysis of transient current signals at the beginning and end. The effectiveness of the method proposed in this article has been fully verified through practical examples.