结合模糊控制的时序数据预测循环神经网络  

Recurrent Neural Networks for Time Series with Fuzzy Control

作　　者：田贤忠[1] 顾思义 胡安娜 TIAN Xian-zhong;GU Si-yi;HU An-na(College of Computer Science and Technology,Zhejiang University of Technology,Hangzhou 310000,China)

机构地区：[1]浙江工业大学计算机科学与技术学院,杭州310000

出　　处：《小型微型计算机系统》2021年第2期241-245,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61672465)资助.

摘　　要：时序数据处理任务中,循环神经网络模型以及相关衍生模型有较好的性能,如长短期记忆模型(LSTM),门限循环单元(GRU)等.模型的记忆层能够保存每个时间步的信息,但是无法高效处理某些领域的时序数据中的非等时间间隔和不规律的数据波动,如金融数据.本文提出了一种基于模糊控制的新型门限循环单元(GRU-Fuzzy)来解决这些问题.本文在GRU的基础上对记忆层增加了一个子空间分解,由模糊控制模块和一个启发式的失效函数组成,根据数据波动和时间间隔决定记忆层保留的信息量,从而提升模型性能.实验表明,相比于其他的循环神经网络模型,在标普500和上证50中选出股票的股价预测任务中,本文提出的模型有较好的表现.In time series data processing tasks,recurrent neural network models and it's derivative models have good performance,such as long short-term memory(LSTM),gated recurrent unit(GRU)and so on.Model's hidden layer can memorize information of every time step,but it unable to handle non-equal interval and irregular data fluctuations in the data efficiently in some fields such as financial data.In this paper,we propose a novel gated recurrent unit based on fuzzy control(GRU-Fuzzy)to solve these problems.We add a subspace decomposition to the hidden layer based on the GRU.It consists of a fuzzy control module and a heuristic failure function.The amount of information retained by the hidden layer is determined according to data fluctuations and time intervals,thereby improving model performance.Empirical studies show that,compared to other recurrent neural network models,our GRU-Fuzzy has state-ofthe-art performance in the stock price prediction tasks,in which the stocks are chosen from S&P 500 and SSE 50.

关 键 词：循环神经网络 时间序列 模糊控制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]