摘要
针对Markov链状态转移预测,提出一种基于离散过程神经元网络(DPNN)的等效状态转移预测方法和模型,探讨DPNN与Markov模型在一定条件下对于系统状态转移特性描述的等价性问题.利用DPNN对时间序列样本的非线性映射机制和自适应学习能力,可通过对时间序列样本集的训练,确定满足Markov链时间序列状态转移约束关系的DPNN,并将其连接权矩阵作为Markov链等效状态转移矩阵.对于任意的Markov链,给出与之等效DPNN的构建方法和在Markov链状态转移概率条件约束下的网络权值矩阵求解算法,仿真实验结果验证了方法的有效性.
Aiming at the problem of Markova chain state transition prediction, this paper presents equivalent prediction method and models based on Discrete Process Neural Networks (DPNN). According to the description of system state transition problems in a certain condition, we discuss the equivalence relation between DPNN and Markov models. Using the non-linear mapping mechanism and the ability in learning adaptively of DPNN against time sequence samples, we train these samples and can determine the DPNN that meets time sequence state transition constraint relations of Markova chain. The connection weight matrix of the DPNN is regarded as the equivalent state transition matrix of Markova chain. For any Markova chain, this paper gives equivalent construction method of DPNN and solving algorithm of network weight matrix under the condition of Markova chain state transition probability. Simulation results prove the effectiveness of this method.
出处
《大庆石油学院学报》
CAS
北大核心
2008年第4期114-117,共4页
Journal of Daqing Petroleum Institute
基金
黑龙江省自然科学基金(ZA2006-11)
黑龙江省科技攻关项目(GZ07A103)
黑龙江省普通高等学校骨干教师创新能力资助计划项目(1055G002)
作者简介
许少华（1962-），男，教授，博士生导师，主要从事模式识别、神经网络、智能信息处理等方面的研究．
