Abstract: Heat exchanger network optimization is an effective way of energy recovery. However, the model with large optimization space for heat exchanger network is often a complex mixed-integer nonlinear programming (MINLP) model with nonlinear and non-convexity constraints, and difficult to get a feasible solution. Based on the stage-wise superstructure, this paper builds a new type of heat exchanger network containing flow split, reflux and non-isothermal mixing. While increasing the optimization space of heat exchange network, linear constraints are set to greatly improve the solvability of MINLP model. Two cases in literature are used to illustrate the contribution of flow split, reflux, isothermal mixing and non-isothermal mixing to the optimization of heat exchanger network, and verify the effectiveness and applicability of the new model.