高级检索

    混合种群多样性自适应遗传操作的HMM训练模型

    HMM Training Model Using Blending Population Diversity Based Aaptive Genetic Algorithm Title

    • 摘要: 针对传统GA方法训练HMM模型所存在的对遗传控制参数具有较强的敏感性的问题:1)提出一种以混合遗传种群多样性为原则的BPD-AGA算法,该算法依据决定种群多样性的基因型和表现型来自适应地调整遗传参数,使整个遗传迭代过程能够在扩大遗传搜索空间的同时提高最优解的质量;2)提出了一种基于BPD-AGA的HMM训练模型,该模型一方面利用BPD-AGA的自适应选择操作,选择对混合种群多样性贡献最大的个体作为竞争优胜者;另一方面利用BPD-AGA的自适应交叉变异操作,在扩大算法搜索空间的同时保护对混合种群多样性贡献较大的个体,从而保证了HMM解个体的全局最优性;3)为了提高BPD-AGA训练的收敛速度,利用Baum-Welch算法从外部对BPD-AGA的收敛性进行了改善,提出了一个从外部和内部同时改善GA性能的BPD-AGA/Baum-Welch混合模型;4)给出了将所提出的模型和算法应用在交通视频车辆行驶状态判别中的实现过程.仿真实验验证了所提出模型和算法的有效性.

       

      Abstract: With the parameters sensibility problem of HMM model trained by GA method, this paper firstly introduces BPD-AGA algorithm based on blending population diversity. In order to make genetic iteration process enlarge the genetic searching space and improve quality of optimal solution, this algorithm adjusts the genetic parameters adaptively by individual genotype and phenotype which determine the population diversity. Secondly, HMM training model based on BPD-AGA has been proposed. This model selects the largest contribution to blending population diversity as the competition champion by BPD-AGA adaptive selection; On the other hand, HMM global optimal solution can be guaranteed by using BPD-AGA adaptive crossover and mutation, and enlarging the genetic searching space to protect the competition champion. Thirdly, this paper mixes Baum-Welch algorithm and BPD-AGA to increase the BPD-AGA training convergence speed, eventually produces a hybrid BPD-AGA/Baum-Welch model that can improve GA from external and internal. Fourthly, the process of the above model and training algorithm applied in traffic video vehicle driving status recognition has been presented. Experiment results prove the effectiveness of the proposed model and training algorithm.

       

    /

    返回文章
    返回