基于量子计算的限制波尔兹曼机网络模型及分类算法  

Net model of restricted boltzmann machine based on quantum computation and its classification method

作　　者：张培林[1] 李胜[1] 吴定海[1] 李兵[2] 周云川[3] 

机构地区：[1]军械工程学院七系,石家庄050003 [2]军械工程学院四系,石家庄050003 [3]军械工程学院军械技术研究所,石家庄050003

出　　处：《振动与冲击》2015年第24期26-31,共6页Journal of Vibration and Shock

基　　金：国家自然科学基金(E51205405);国家自然科学基金(E51305454)

摘　　要：为进一步简化模型结构,提高模式识别性能,提出一种基于量子计算的限制波尔兹曼机网络模型(Restricted Boltzmann Machine Based on Quantum Computation,QRBM)。在QRBM网络中,依据RBM的网络结构,以量子计算为基础。首先,对数据进行量子化编码。然后,执行量子操作,生成网络的权值矩阵以简化步骤、提高计算效率。之后,确定网络层数以提高准确率,缩短执行时间。最后,实现QRBM模型参数的更新,从而达到故障分类的目的。将该方法用于齿轮箱模式识别中,提取齿轮箱的正常、齿面磨损、齿根裂纹和断齿等振动信号的数据作为原始特征,采用QRBM神经网络模型进行模式识别。实验结果表明,QRBM分类算法在分类准确率和执行时间上获得的效果比普通神经网络、支持向量机和RBM网络更好,验证了本文方法的有效性和可行性。In order to simplify the structure of model and enhance the performance of pattern recognition,a net model of restricted Boltzmann machine based on quantum computation（ QRBM） was proposed. In the QRBM network,based on the net structure of RBM and quantum computation,the data were coded with quantum states. Then,by quantum operation,a weight matrix was created for simplifying computation step and enhancing computation efficiency.After that,the number of net layers was determined to improve accuracy and shorten execution time. Finally,the parameters in the model were updated. The method has been applied in gear fault diagnosis. The data extracted form vibration signals of a gear box under the conditions of normal states,wearing,crack and breakage were taken as the original features and the. QRBM was used for diagnosis with the feature set. The results indicate that,compared with the methods of neural network,SVM and RBM network,the QRBM has better performance in classification accuracy and execution time. The efficientcy and feasibility of the method was proved.

关 键 词：量子计算 限制波尔兹曼机 神经网络 齿轮 模式识别 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]