Rapid and accurate assessment of power network loss in the power system has become a key research topic for the vast and diverse dataset of power grid operation. This study integrates data mining techniques with typical scenario modeling concepts and innovatively designs a distribution area power network loss rate multi population parallel spectral clustering evaluation strategy that incorporates distribution characteristics. Firstly, clustering attributes are determined for power network loss evaluation, and a power network loss evaluation framework based on clustering algorithms is proposed. Based on power flow calculation, the distribution characteristics and indicator system of each node’s output are analyzed; Secondly, in order to improve the clustering accuracy of power network loss evaluation, spectral clustering algorithm is introduced, and automatic algorithm design is carried out to address the issue of manually setting the initial number of clusters and cluster centers. Then, multi cluster partitioning and parallel computing methods are used to significantly improve the computational efficiency of spectral clustering algorithm; Finally, to verify the practicality of this method, a provincial power grid was selected as a case study. The results showed that this method not only has high accuracy in evaluating power network loss, but also has excellent computational efficiency, demonstrating good feasibility in practical engineering applications.