基于改进Transformer模型的多元时间序列预测
Multi-variate time series forecasting based on improved Transformer model
程艺锐 1李果1
作者信息
- 1. 南阳师范学院人工智能与软件工程学院,河南南阳 473061
- 折叠
摘要
在无线数据传输中,环境干扰和网络拥塞导致的数据丢包和缺失问题显著影响了时间序列预测的稳定性.为了解决这个问题,提出了一种名为TFKNet的时间序列预测模型.该模型基于Transformer的多维时间序列数据预测方法,在传统Transformer模型的基础上,结合时间卷积网络(TCN)和傅立叶频率特征提取技术增强了模型对局部特征的捕捉能力和多频率特征的提取能力,引入Kernel Attention Networks(KAN)提高了模型的预测性能.实验结果表明,与Transformer、Informer、Reformer、Autoformer传统方法相比,TFKNet模型在时间序列长预测任务中预测误差MAE分别平均降低0.052 2、0.111 7、0.120 9、0.192 2.
Abstract
In wireless data transmission,issues such as packet loss and missing data due to environmental inter-ference and network congestion significantly affect the stability of time series prediction.To address this problem,a time series prediction model called TFKNet is proposed.This model is based on the Transformer method for multidimensional time series data prediction.It integrates Temporal Convolutional Networks(TCN)and Fourier frequency feature extraction techniques into the traditional Transformer model,enhancing the model's ability to capture local features and extract multi-frequency features.Additionally,Kernel Attention Networks(KAN)are introduced to improve the model's predictive performance.Experimental results demonstrate that,compared to traditional methods such as Transformer,Informer,Reformer,and Autoformer,the TFKNet model achieves an average reduction in prediction error(MAE)by 0.052 2,0.111 7,0.120 9,and 0.192 2,respectively,in long-term time series prediction tasks.
关键词
数据预测/Transformer/TCN/KAN/傅立叶频率特征Key words
data prediction/Transformer/TCN/KAN/Fourier frequency features引用本文复制引用
出版年
2025
国家科技期刊平台
NSTL
万方数据