基于改进遗传算法的DBN结构自适应学习算法  

DBN Structure Adaptive Learning Algorithm Based on Improved Genetic Algorithms

作　　者：孙美婷 刘彬[2] SUN Mei-ting;LIU Bin(School of artificial intelligence and automation,Ministry of Information Science,Beijing University of Technology,Beijing 100124,China;School of College of Information Science and Engineering,Yanshan University,Qinhuangdao,Hebei 066004,China)

机构地区：[1]北京工业大学信息学部人工智能与自动化学院,北京100124 [2]燕山大学信息科学与工程学院,河北秦皇岛066004

出　　处：《计量学报》2021年第1期91-99,共9页Acta Metrologica Sinica

摘　　要：针对动态贝叶斯网络(dynamic bayesian network,DBN)是NP困难问题,提出基于改进遗传算法的DBN结构自适应学习算法。该算法计算最大互信息和时序互信息完成DBN结构搜索空间的初始化。在此基础上设计改进遗传算法,引入评分标准差构建交叉概率和变异概率的自适应调节函数,以降低结构学习过程陷入局部最优解的概率。仿真结果表明,该算法在无先验知识的情况下,相比其他优化算法,汉明距离和运行时长平均减少了30%,37.3%,评分值平均增大了18.0%。Aiming at the NP-hardness problem of dynamic bayesian network,dynamic bayesian structure adaptive learning algorithm based on improved Genetic Algorithm is proposed. The maximum mutual information and timing mutual information are first used in the proposed algorithm to build initial structure,completing the initialization of the search space for DBN structures. Based on this,an improved genetic algorithm is presented. An adaptive control function of crossover probability and mutation probability is constructed introducing the grading standard deviation in order to reduce the probability of getting trapped in a local optimum. Compared with other optimization algorithm,experimental results indicates that the IMGA-DBN algorithm can significantly decrease nearly 30% and 37% in the hamming distance and running time separately. Meanwhile,IMGA-DBN increase 18. 0% in the scoreing metric values without prior information.

关 键 词：计量学 动态贝叶斯网络 时序互信息 得分标准差 自适应学习 

分 类 号：TB973[一般工业技术—计量学]