摘要
本文将模拟退火组合优化法的搜索控制过程与最优解的更新过程加以分离,提出了一种有效的改进方法.它得到的解总是优于(至少不劣于)原有模拟退火法的解,尤其是在因有关参数选择不合适使原模拟退火法获得的解较差时,改进法仍能得到较满意的解.另外,为减少计算量,改进法还提供了一种判断 Metropolis 抽样停止和退火过程结束的有效方案.最后,给出了改进法在两个模式识别典型问题上的应用,以及计算机实验结果.
Through separating the procedure of search controlling from that of solution updating,an improved versionof simulated annealing method is proposed for combinatorial optimization.The solution obtained by the improvedmethod is always superior(at least not inferior)to that by the original simulated annealing method,especially inthe case that the parameters chosen for simulated annealing are not appropriate,the original method usually givesworse solutions.However,the improved method can still supply satisfying solutions.The improved version alsoprovided an effective way for deciding when Metropolis sampling is to be finished and where the annealing proce-dure is to be stopped.Furthermore,the improved method has also been used to solve two typical pattern recogni-tion problems.The computer experiment results have shown the effectiveness of the improved method.
出处
《信息与控制》
CSCD
北大核心
1990年第3期1-7,共7页
Information and Control
关键词
模拟退火法
组合优化法
聚类分析
simulated annealing
combinational optimization
attributed graph matching
clustering analysis
