Graph-theoretic System Decomposition Based on Modified Cosine-based Similarity

A graph-theoretic system decomposition method based on modified cosine-based similarity is proposed to solve the current problem of system decomposition of industrial processes. The correlations between input variables and output variables are analyzed, the initial subgraphs are built according to their weights, and the smallest subgraphs are selected for trial fusion with other subgraphs. The trial fusion result with the largest fusion index is selected as the fusion result, and the termination judgment is made. The system decomposition is more accurate than the graph-theoretic system decomposition with Pearson's correlation coefficient. Finally, the validity is verified by the Shell heavy oil fractionation column.