论文部分内容阅读
针对传统模拟退火算法存在收敛速度慢、执行时间长的缺点,本研究提出了一种并行在线的模拟退火算法及其优化策略,并将其运用到森林景观分类中。研究人员运用多马尔科夫链异步通信和同步通信两种策略实现模拟退火算法的并行处理。在Solomon提供的标准测试集上对并行算法性能进行测试和分析,得出并行算法时线程间的通信可以提高目标解的搜索效率。与此同时,同步通信策略目标解的搜索效率优于异步通信策略,但是会增加一些通信负载的成本。通过大量实验得出森林分类经营代价与线程沟通周期、链长和线程数目的关系,从而节省景观分类的时间代价,进而解决一些NP难题。
Aiming at the shortcomings of slow convergence speed and long execution time of traditional simulated annealing algorithm, this paper presents a parallel on-line simulated annealing algorithm and its optimization strategy, which is applied to forest landscape classification. Researchers use the multi-Markov chain asynchronous communication and synchronous communication two strategies to achieve parallel processing of simulated annealing algorithm. In the standard test set provided by Solomon, the performance of parallel algorithms is tested and analyzed. It concludes that the communication between threads in parallel algorithm can improve the search efficiency of the target solution. At the same time, the search efficiency of the target solution of synchronous communication strategy is better than asynchronous communication strategy, but it will increase the cost of some communication load. Through a large number of experiments, the relationship between the management cost of forest classification and the thread communication cycle, chain length and number of threads is obtained, saving the time cost of landscape classification and then solving some NP problems.