In optimizing heat exchanger networks by random walk algorithm with compulsive evolution (RWCE), individual structures become similar, which leads to a decline in population diversity and algorithmic global search capability. It becames difficult to optimize the structure of heat exchanger networks further. To explore similarity of individual structures, two evaluation indexes were established to measure similarity level of individual structures.It was found that the number of similar heat units in individual structure increases gradually and the difference of heat load between similar heat units decreases during the optimization process which resulting the evolution of individual structures in the same direction. Therefore, taking the improvement of population diversity as guiding direction, a repulsive-evolution strategy for heat units was proposed. It changes the direction of structure evolution and reduces the scale of similar structure through increasing the heat load gap of similar heat exchange units and promoting differential evolution of similar heat exchange units.Finally, 20SP and 15SP were used to verify effectiveness of the strategy. Total annual cost of the structure saved 12 105 ＄·a^-1 and 52 535 ＄·a^-1, respectively, compared with the optimal result in literature, which indicating that the strategy enhances the algorithmic global search capability effectively.