混合训练的DHMM及其在发射机状态检测中的应用
doi: 10.3724/SP.J.1146.2006.01875
Hybrid Training DHMM and Its Application to Check Transmitter Power
-
摘要: 隐马尔可夫模型(HMM)是一种双随机过程，其训练方法B-W算法是一种基于爬山算法，容易陷入局部最优且对初始参数值依赖性大。为了提高模型的有效性，该文提出了将改进的模拟退火(SA)算法和B-W算法相结合的混合训练方法，解决了受模型参数初值影响的问题并能实现全局搜索。将其应用于发射机功率状态检测中，实验结果证明该方法准确性高，收敛速度快和稳定性好，是一种很有实用价值的新方法。Abstract: HMM model is a double random processing which is trained with B-W algorithm, this algorithm based on hill-climbing is easy to lead to locally optimal solutions and its validity is greatly depend on model initial parameters. In order to improve the validity of model, this paper proposes a hybrid training method which combine the B-W algorithm with improved SA algorithm. With this hybrid method the validity of model is not influenced by the model initial parameters and the global optimal solution can be easily gained. Applying this hybrid method to check transmitter power, the experimental results show that proposed method is practical method and own qualities such as high veracity, rapid converged and good stability.
-
Rabiner L R. A tutorial on hidden Markov models andselected applications in speech recognition [J].Proc. IEEE.1989, 77(2):257-286[2]Jie Ying, and Kirubarajan T. A hidden Markov-basedalgorithm for fault diagnosis with partial and imperfect tests[J].IEEE Trans. on System Man and Cybernetics.2000, 30(4):463-473[3]Gales M J F. Cluster adaptive training of hidden Markovmodel [J].IEEE Trans. on Speech and Audio Processing.2000,8(4):417-432[4]茅晓泉, 胡光锐. 一种基于梯度的HMM 参数重估方法[J]. 上海交通大学学报, 2002, 36(5): 683-685.[5]Chau C W, Wang S K, and Diu C K. Optimization of HMMby a genetic algorithm [C]. Proc, Munich, Germany, 1997:1727-1730.[6]何志国, 何钦铭, 陈奇. 语音识别中进化计算与MDI 的HMM训练方法 [J].系统工程理论与实践, 2006, 26(7): 54-58.[7]孙延奎, 朱小燕. 人工智能[M]. 第一版. 北京: 清华大学出版社, 2004: 220-246. -
计量
- 文章访问数: 2942
- HTML全文浏览量: 87
- PDF下载量: 818
- 被引次数: 0
下载:
