Abstract: As nodes-based nonstructural model with stream splits (NNM-SS) applied in heat exchanger network synthesis, optimization may be trapped in local optima due to the difficult in generating suitable structures with stream splits. Structural perturbation strategy is proposed and is added into random walk algorithm with compulsive evolution (RWCE-SS). During optimization, a heat exchanger without stream splits is chosen under a certain probability, and then generating a new substream structure on its position to add the number of stream splits in the network. Optimization results obtained by retrofitted algorithm are better than those in literatures.