In the process of optimizing the mass exchange network by random walk algorithm with compulsive evolution, some individuals are in a long-term disadvantaged state in the competition and the individual structure is highly similar. RWCE algorithm with individual elimination is proposed to optimize the mass exchange network. That is, in a certain period, through the real-time monitoring of the optimal state of the individuals in the population, the network structure of the individual is first standardized, so as to identify the similar quality exchangers in the structure, and then evaluate the structural similarity of the individual according to its number. Divide the individuals in the population into several groups, and use the total annual cost as an indicator to evaluate the individual optimization performance, and eliminate the disadvantaged and similar individuals in the population. Strengthen the information exchange of individuals between populations, improve individual optimization vitality and population diversity, and effectively enhance the global search ability of the algorithm to optimize the mass exchange network. Applying this method to two examples of mass exchange networks, the optimization results are better than the best results in the literature, indicating that the method can change the evolution direction of the structure, stimulate the differential evolution of individuals between populations and maintain the optimization vitality of individuals, and effectively improve the global optimization performance of the algorithm.

As RWCE algorithm is used to optimize heat exchanger networks, the existence of fixed investment cost makes it difficult to produce and retain heat exchanger units with small loads. It makes structures that have not yet fully evolved be eliminated too early. We established a relaxation method of fixed investment cost. The fixed investment cost and the heat exchange of heat exchange unit are processed according to their variation. With heat exchange of heat exchange unit, relaxation strength in optimization process is adjusted in real time to guide and promote the smooth generation and evolution of small load heat exchange unit, so as to enhance structural evolution ability of heat exchange network. The relaxation strategy is applied to 9SP and 15SP examples to verify that the relaxation strategy promotes structural evolution effectively and improves optimization quality. The optimal results (2 903 528 ＄ ·a^-1, 5 115 061 ＄ ·a^-1) are better than those in literatures.

An evolutionary algorithm of multi-directional compulsive search strategy for dominant individuals is proposed to evaluate the health of global search of current population by examining individual difference index.When the optimization results stagnate for a long time, the health index becomes worse and the multi-direction forced search strategy of the dominant individual is started to expand the search direction of the dominant individual near the local optimal solution region, so as to ensure the global search ability of the algorithm.The strategy was applied to 15SP and 20SP calculation examples and results were compared with optimal results in the literature.They are decreased by 1.09% and 0.83%, respectively.It in dicates that the dominant individual multi-directional compulsive search strategy exerts fully the evolutionary potential of the dominant individual and improves the optimization efficiency of the algorithm.

As random walk algorithm with compulsive evolution is used to optimize heat exchanger networks, it promotes effectively structural evolution by accepting difference solution with certain probability. Non-greedy difference solution search mechanism is the key for the target function to jump out of local extremum. We analyze accept imperfect networks mechanism, explore different sources of imperfect networks of the optimization phase structure, and show the next year comprehensive cost change. We found that most acceptable solution return to the original structure in the subsequent optimization. There exist a lot of invalid parameter passing process. In this paper we present a selective acceptance of imperfect networks strategy, in which only difference solutions with new individuals are accpted based on an exponential increasing probability formula. Structure of new individual is protected to enhance its survival rate in structure. Finally, 9SP₂ and 20SP examples were used for verification. The optimal results are better than those in literature, which shows feasibility of the strategy.