基于聚类和神经网络的多变量时间序列预测方法
首发时间:2019-11-29
摘要:多变量时间序列数据在日常生活中分布广泛,并且在很多领域中得到广泛研究。但是其中仍然存在着一些挑战,比如提取时间序列内部的短期和长期依存关系、捕获序列中的多种周期模式。本文提出了聚类卷积递归神经网络来应对其中的种种问题。经过多变量时间序列聚类之后,模型分别使用卷积递归神经网络与展平卷积递归神经网络来提取不同聚类类别之间与类别内部的差异性与相似性,并且组件中同时使用注意力机制和多个卷积层来捕获序列中不同的周期性模式。本文中提出的模型在各种数据集上的效果都取得了显着改善且具有很高的实用性,所有实验结果将在稍后的论文中展示。
关键词: 人工智能 多变量时间序列 注意力 卷积网络 递归网络
For information in English, please click here
A Clustering and Neural Network Based Model for Multivariate Time-Series Prediction
Abstract:Multivariate time series are widely distributed in our daily life. And multivariate time series prediction is extensively studied with ubiquitous applications across many domains. But ther are still challenges in it, like extracting short- and long-term dependency, capturing different periodic pattern. This paper proposes a novel deep learning framework named Clustering Convolutional Recurrent Neural Network. Starting with time series clustering, Convolutional Recurrent Neural Network and Flatten Convolutional Recurrent Neural Network extract differences and similarities among clusters and variables. This model also use attention mechanism and multiple convolution layers to capture different periodic patterns. This model has achieved significant improvement in diverse datasets and has great practicabilities, all experimental results will be shown through our paper later.
Keywords: Artificial intelligence Multivariate time series Attention Convolutional neural network Recurrent neural network
引用
No.****
动态公开评议
共计0人参与
勘误表
基于聚类和神经网络的多变量时间序列预测方法
评论
全部评论0/1000