Hydrogen is a crucial and expensive resource for refineries. Hydrogen network integration is an effective way to conserve fresh hydrogen. On the basis of relative concentration analysis, a mixed integer linear programming (MILP) mathematical model is established to synthesize refinery multi-impurity hydrogen networks. By correlating the hydrogen consumption of a sink with its oil processing throughputs, this model reveals the fresh hydrogen variation tendency caused by fluctuation of each capacity, and determines the key sink(s) of a hydrogen network with multiple impurities on the premise of minimum fresh hydrogen consumption. With a structure matrix variable, practical hydrogen networks can be analyzed through the proposed model. The results show that the novel method in this paper is superior to traditional absolute concentration-based methods in fresh hydrogen conservation under cases of static, fluctuant throughputs, and fluctuant throughputs with restricted network structure. Two cases are employed to demonstrate its application.