论文部分内容阅读
针对贝叶斯优化算法(BOA)中学习贝叶斯网络结构时间复杂度较高的问题,提出了一种可以快速收敛的基于K2的贝叶斯优化算法(K2-BOA).为了提升收敛速度,在学习贝叶斯网络结构的步骤中进行了2处改进:首先,随机生成n个变量的拓扑排序,加大了算法的随机性;其次,在排序的基础上利用K2算法学习贝叶斯网络结构,减少了整个算法的时间复杂度.针对3个标准Benchmark函数的仿真实验表明:采用K2-BOA算法和BOA算法解决简单分解函数问题时,寻找到最优值的适应度函数评价次数几乎相同,但是每次迭代K2-BOA算法运行速度提升明显;当解决比较复杂的6阶双极欺骗函数问题时,K2-BOA算法无论是运行时间还是适应度函数评价次数,都远小于BOA算法.
In order to improve the time complexity of learning Bayesian network in Bayesian optimization algorithm (BOA), a K2-based Bayesian optimization algorithm (K2-BOA) that can converge rapidly is proposed. In order to improve convergence speed , Two improvements are made in the steps of learning Bayesian network structure. Firstly, randomly generating the topological ordering of n variables, increasing the randomness of the algorithm. Secondly, using the K2 algorithm to learn Bayesian Network structure reduces the time complexity of the whole algorithm.According to the simulation results of three standard Benchmark functions, it shows that when using the K2-BOA algorithm and the BOA algorithm to solve the simple decomposition function, the number of the fitness function to find the optimal value is almost However, the K2-BOA algorithm improves significantly with each iteration. When solving the complicated 6-order bipolar cheat function problem, the K2-BOA algorithm is far less than the BOA algorithm in terms of running time and fitness function evaluation.