RWCE优化换热网络的不可行解影响分析及强化策略

换热网络优化问题常以外罚函数法处理约束，赋予违反约束的不可行解较大的罚值。强制进化随机游走算法(RWCE)优化换热网络时，其非贪婪搜索机制使不可行解以一定概率被保留，从而改变全局寻优过程。本文首先分析不可行解对优化进程的影响，揭示偏移量较小的不可行解对结构进化的促进作用；然后提出差解概率动态调整策略，合理利用不可行解的正面作用，强化结构进化能力；最后，鉴于上述优化结果中偏移量较小的不可行解居多，提出一种可行化策略，通过分段罚指数和双种群优化技术促使过程中有潜力的不可行解快速返回可行域，并提升优化质量。将结合两条强化策略的改进算法应用于16股流与15股流算例，优化结果较文献最优解分别节省了0.35%、0.48%，表明改进后的算法较原算法全局搜索能力得到了显著提升。

Influence analysis and enhancement strategy of infeasible solutions for heat exchanger network optimization with RWCE

In optimization of heat exchanger network (HEN), the exterior penalty method was usually used to deal with the constraints, which gave a large penalty value to the infeasible solutions that violated the constraints. When random walk algorithm with compulsive evolution (RWCE) was applied to the HEN optimization, the infeasible solutions maybe accepted with a certain probability due to the non-greedy search mechanism of RWCE, thus changing the global optimization process. In this paper, the influence of infeasible solutions on the optimization process was analyzed firstly, revealing that the infeasible solutions with small offset could promote structure evolution. Then, a dynamic adjustment strategy of the acceptance probability of bad solutions was proposed aiming to reasonably utilize the positive effect of the infeasible solutions to enhance the structure evolution. Finally, considering that most of the above optimization results were infeasible solutions with relatively small offset, a feasible strategy was proposed. Through the application of the piecewise penalty index technique and two-population optimization technique, the infeasible solutions with evolution potential were enabled to rapidly return to the feasible regions, the optimization quality could also be improved. The modified algorithm combining the two reinforcement strategies and RWCE was applied to the cases involving 16 streams and 15 streams, whose results respectively saved 0.35% and 0.48% as compared to the optimums in literature, indicating that the global search ability of the modified algorithm was significantly improved compared with the original algorithm.