Multi-Task Learning for Time Series Forecasting Using NARMAX-LSTM
Multi-Task Learning for Time Series Forecasting Using NARMAX-LSTM
复制标题
DOI：
10.1109/icac55051.2022.9911071
复制
发表时间：
2022-09
期刊：
影响因子：
--
通讯作者：
N. Nikentari;Hua‐Liang Wei
中科院分区：
文献类型：
--
作者：
N. Nikentari;Hua‐Liang Wei
传统的基于神经网络的时间序列预测模型通常是针对单个任务设计的，忽略了其他相关任务的知识或信息来源。另一方面，多任务学习(MTL)提供了一个新的视角，即一个模型不仅适用于一个预测案例或任务，而且可以同时适用于多个任务。本文提出的MTL模型是基于具有外生输入的非线性自回归滑动平均(NARMAX)模型和长-短期记忆(LSTM)神经网络。使用三个不同的时间范围和不同的案例/任务对NARMAX-LSTM模型进行了测试和验证。将该模型与其他三种线性和非线性模型的预测性能进行了比较。比较结果表明，多任务NARMAX-LSTM模型不仅具有更好的预测性能，而且避免了负迁移和过拟合。
Traditional time series forecasting models based on neural networks are usually designed for a single task, which ignores the source of knowledge or information from other related tasks. On the other hand, multitask learning (MTL) offers a new perspective where one model is not only for one forecasting case or task but several tasks at the same time. The MTL model proposed in this study is based on the Nonlinear AutoRegressive Moving Average with eXogenous input (NARMAX) model and Long-Short Term Memory (LSTM) Neural Networks. The NARMAX-LSTM model is tested and validated using three different time horizons and different cases/tasks. The model forecasting performance is compared with that of three other linear and nonlinear models. The comparison results show that the multi-task NARMAX-LSTM model does not only have better prediction performance but also avoid suffering from negative transfer and overfitting.