改进的自适应模拟退火算法及其在过程综合中的应用  

Improved Adaptive Simulated Annealing Algorithm for MINLP Problems in Process Synthesis

作　　者：杜红彬[1] 薛东峰[2] 姚平经[1] 

机构地区：[1]大连理工大学化工系统工程研究所,辽宁大连116012 [2]清华大学化工系统工程研究所,北京100084

出　　处：《高校化学工程学报》2002年第1期106-110,共5页Journal of Chemical Engineering of Chinese Universities

摘　　要：为有效解决化工过程综合中的MINLP问题,针对连续变量的模拟退火算法搜索慢的缺点,提出了一种改进的自适应模拟退火算法 (Adaptive Simulated Algorithms, ASA),采取自适应调整温度和搜索步长两种策略, 大大加快搜索速度,提高最优解的质量。实算结果充分体现了所提出算法的优点,并很好地应用于化工过程综合问题。Improved Adaptive Simulated Annealing Algorithm (IASA) suitable for the optimization of mixed-integer non-linear programming(MINLP) problems was presented. Since these problems have irregular, large and non-convex solution space, many methods fail to find the global optimal. Simulated Annealing Algorithm (SA), as a kind of random methods, is thought efficient to deal with this puzzle. But as most random algorithms search the optimal solution in an asymptotic way, SA may have low computing efficiency and the probability of being trapped in the local area especially when the scale of the problem is very large. The presented IASA has two advantages in improving the performance of the traditional SA. The first one is that the new algorithm adopted adaptive strategy of adjusting the annealing temperature to improve computing speed. Before every time temperature is decreased, the proportion of the number of reacceptance in refusal in the trial solutions is calculated in the inner loop of IASA to decide which kind of annealing ratios should be adopted. The second one is that IASA chooses a proper searching step of each variable adaptively by the relation between the number of acceptance and the number of refusal in the trial solutions. The principle of the two strategies was theoretically analyzed. In the test problems, the proposed algorithm showed better performance both in speed and solution quality compared with traditional SA and genetic Algorithm. To solve the MINLP problems, the integer variables were expressed by rounding the corresponding continuous variables, and the constraint conditions can be got rid of by penalizing the objective function. At last, we apply the new algorithm to some MINLP cases successfully.

关 键 词：MINLP 过程系统综合 自适应模拟退火 优化 化工过程综合 

分 类 号：TQ021.8[化学工程] TQ015.9