基于时间序列的前馈神经网络的研究  

Research on feedforward Neural Network based on time series

作　　者：彭一帆 PENG Yifan(Wuhan Research Institute of Posts and Telecommunications,Wuhan 430034,China)

机构地区：[1]武汉邮电科学研究院,湖北武汉430034

出　　处：《电子设计工程》2021年第4期102-106,111,共6页Electronic Design Engineering

摘　　要：针对时间序列识别的难点,介绍了一种被称为动态时间规整神经网络(DTW-NN)的新颖的时间序列识别模型,DTW-NN是一种利用动态时间规整(DTW)的弹性匹配能力来将层的输入与权值动态对齐的前馈神经网络。通过这种方式,DTW-NN能够解决时间序列识别的困难,例如在前馈结构中的时间失真和可变模式长度。结合在4个不同的数据集上的实验,证明了DTW-NN的有效性:在线手写字符、基于加速度计的活跃的日常生活活动、阿拉伯数字口语的Mel频率倒谱系数(MFCC)和叶形的一维质心半径序列。通过在这些数据集上获得的结果,证明了该方法是一种有效的时间模式学习的通用方法。Due to the difficulties with time series recognition,this paper describes a novel model for time series recognition called a Dynamic Time Warping Neural Network Dynamic Time Warping(DTW-NN).DTW-NN is a feedforward neural network that exploits the elastic matching ability of(DTW)to dynamically align the inputs of a layer to the weights.In this way,the DTW-NN is able to tackle difficulties with time series recognition such as temporal distortions and variable pattern length within a feedforward architecture.Experiments on four distinct datasets demonstrate the effectiveness of DTW-NNs:online handwritten characters,accelerometer-based active daily life activities,spoken Arabic numeral Mel-Frequency Cepstrum Coefficients(MFCC),and onedimensional centroid-radii sequences from leaf shapes.It shows that the proposed method is an effective general approach to temporal pattern learning by achieving state-of-the-art results on these datasets.

关 键 词：神经网络 动态时间规整 时间内核 时间序列 动态编程 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]