摘要
多维时间序列数据存在于实际生活中,包括楼盘价格、道路上交通流量、不同区域的CO_(2)浓度等等。循环神经网络(RNN)是有效处理时间序列数据的一种模型,其变体长短期记忆网络(LSTM)有效解决了RNN反向传播路径过长、易产生梯度爆炸或消失的问题。以四元数代替实数进行网络参数传播,通过四元数内部结构的依赖性,捕获多维时间序列特征之间的内部关系,使得多维时间序列特征中固有的结构信息得到很好的保存。
Multidimensional time series data exists in real life,including property prices,road traffic flow,CO_(2) concentrations in different regions,and so on.Cyclic neural network(RNN)is a model for effectively processing time series data.Its variant long short-term memory network(LSTM)effectively solves the problem that the reverse propagation path of RNN is too long,which is easy to cause gradient explosion or disappearance.This paper uses quaternions instead of real numbers for network parameter propagation,capturing the internal relationships between multidimensional time series features through the dependence of the internal structure of quaternions,so that the inherent structural information in multidimensional time series features is well preserved.
出处
《工业控制计算机》
2024年第2期129-130,共2页
Industrial Control Computer
