Abstract: In order to analyse the actual energy consumption of the cement manufacturing process, an improved DBSCAN (density-based spatial clustering of applications with noise) algorithm is proposed, and the actual energy consumption benchmark of the cement manufacturing process is established based on the linear minimum variance fusion algorithm. In large time scales, the clustering number of energy consumption data can not be obtained directly because of the variation of operating conditions and noise data. To solve this problem, an improved DBSCAN algorithm is applied to cluster the energy consumption data of each process unit in the cement manufacturing process. Classifications, clustering centers and their variances of energy consumption variables are obtained. The actual energy consumption benchmark of each process unit that contains comprehensive factors is obtained by using the clustering centers and their variances of each energy variable with the linear minimum variance fusion algorithm. The application case proves that the improved DBSCAN algorithm can effectively reduce the query number of core objects and the execution time of clustering. The energy efficiency benchmarking obtained by data fusion can reasonably reflect the actual energy efficiency of the cement manufacturing process, and reveal energy saving potentialities.