摘要
贝叶斯网络理论在人工智能领域发挥着重要作用。贝叶斯网络从数据中学习知识的能力使得它在医学、故障诊断、预测等领域的应用迅速发展起来。结构学习算法成为贝叶斯网络的重要研究方向,它能够有效分析变量之间依赖关系,合理挖掘数据和知识。K2算法评分性能突出,而爬山算法能有效弥补K2评分法的解空间过于复杂的问题。论文结合K2评分函数和爬山策略,提出了K2&HC算法。同时,K2&HC算法在爬山策略中融入了回溯原理,解决了贝叶斯结构学习算法中存在的收敛于局部最优的问题,合理优化了算法的性能。同K2和K2SA算法进行仿真对比,得出在精度和收敛速度综合性能上K2&HC表现突出的结论。
Bayesian network plays an important role in the field of artificial intelligence .The capability of learning knowledge from data makes it develop rapidly in medicine ,fault diagnosis ,forecasting and other fields .Structure learning al-gorithm of Bayesian network becomes an important research area ,which can effectively analyze dependencies between varia-bles and discover knowledge and data properly .Hill-Climbing strategy can reduce the complex solution space and improve the performance of structure learning algorithm .At the same time ,the K2 algorithm is outstanding on the performance of sco-ring .Combined the scoring function of K2 with the efficient Hill-Climbing strategy ,the K2＆HC algorithm is proposed . Meanwhile ,Backtracking principle integrates into the search strategy to solve the problem about the structure learning algo-rithm converging to a local optimum ,which can optimize the performance of the algorithm .Contrast with K2 and K2SA sim-ulation ,the conclusions are made that K2＆HC algorithm is outstanding on the comprehensive performance of the accuracy and convergence rate .
出处
《计算机与数字工程》
2014年第7期1137-1140,1145,共5页
Computer & Digital Engineering
关键词
K2算法
爬山法
评分搜索
贝叶斯网络
结构学习
K2 algorithm
hill-climbing
search-and-score
Bayesian network
structure learning
