基于自注意力TCN模型的人民币汇率预测  

RMB Exchange Rate Prediction Based on Self-attention TCN Model

作　　者：阳芊芊 YANG Qianqian(School of Business,Zhengzhou University,Zhengzhou 450001,China)

机构地区：[1]郑州大学商学院,郑州450001

出　　处：《科技和产业》2023年第8期138-142,共5页Science Technology and Industry

摘　　要：随着人民币汇率市场化改革的持续推进,汇率的潜在波动空间扩大。汇率波动关系到投资策略的制定及风险监管的实施,对其进行准确预测具有重要意义。深度学习以其强大的非线性拟合及特征提取能力在汇率预测领域受到关注。选取TCN(时序卷积网络)模型并在其基础上加入自注意力机制进行改进,构建SA-TCN模型,同时加入VIX(波动率指数)衡量市场情绪,以实现对汇率波动的精准预测。实证结果显示,相较于目前常用的LSTM、GRU模型及基础的TCN模型,SA-TCN模型预测效果最佳。As the market-oriented reform of the RMB exchange rate continues to advance,the potential fluctuation space of the exchange rate will expand.Exchange rate fluctuation is related to the formulation of investment strategy and the implementation of risk supervision,so it is of great significance to accurately forecast it.Deep learning has attracted attention in the field of exchange rate prediction because of its powerful nonlinear fitting and feature extraction ability.The TCN(temporal convolutional networks)model is selected and improved by adding self-attention mechanism,the SA-TCN model is built,and the VIX(volatility index)is added to measure market sentiment,and achieve accurate prediction of exchange rate fluctuations.The empirical results show that,compared with the commonly used LSTM,GRU model and the basic TCN model,the SA-TCN model has the best prediction performance.

关 键 词：汇率预测 时序卷积网络(TCN) 自注意力机制 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] F832.6[自动化与计算机技术—控制科学与工程]