In view of structural evolution difficulty caused by integer variables in heat exchanger network with fixed investment cost, a fixed investment cost relaxation strategy is proposed. It simplifies mathematical model of the problem by relaxing fixed investment cost. As heat exchanger is very small, fixed investment cost is almost zero. With increase of heat exchanger, fixed investment cost is gradually increased with a certain slope, and finally equals to the actual value. Controlling the change slope by the relaxation strength coefficient, on the basis of ensuring reliability of optimization results, heat exchangers with structural evolutionary obstacles are guided to generate or eliminate. The strategy is used to two cases in the literature, and effect of this strategy on generation or elimination of heat exchanger under different relaxation strength is investigated. Finally, a random walk algorithm with compulsive evolution based on fixed investment cost relaxation strategy is proposed. The algorithm is applied to a heat transfer network case. The result is superior to existing literature.

Aiming at popularing diversity is disappared or other reasons in the later stage of optimiging heat exchanger network synthesis problem, it is difficult to find direction of evolution which makes total annual cost further reduce in optimization. A coupling evolution strategy of heat exchanger was proposed. In later stage of heuristic algorithm optimization, heat exchangers with no heat load are taken into coupled evolution by a certain probability distribution, to find coupling match to reduce cost. It shows that the strategy is effective. The strategy was combined with RWCE algorithm to form a hybrid algorithm. Firstly, RWCE algorithm was used to explore solution domain to find potential solutions by its strong global searching ability. Secondly, the coupled evolution strategy is applied to further optimize for these explored solutions. Thirdly, the further optimized solutions are fed back to RWCE algorithm after mutated. The hybrid algorithm is applied to 10SP2 and 15SP, and better optimization results are obtained.