A Novel Encoder-Decoder Model for Multivariate Time Series Forecasting

Abstract

The time series is a kind of complex structure data, which contains some special characteristics such as high dimension, dynamic, and high noise. Moreover, multivariate time series (MTS) has become a crucial study in data mining. The MTS utilizes the historical data to forecast its variation trend and has turned into one of the hotspots. In the era of rapid information development and big data, accurate prediction of MTS has attracted much attention. In this paper, a novel deep learning architecture based on the encoder-decoder framework is proposed for MTS forecasting. In this architecture, firstly, the gated recurrent unit (GRU) is taken as the main unit structure of both the procedures in encoding and decoding to extract the useful successive feature information. Then, different from the existing models, the attention mechanism (AM) is introduced to exploit the importance of different historical data for reconstruction at the decoding stage. Meanwhile, feature reuse is realized by skip connections based on the residual network for alleviating the influence of previous features on data reconstruction. Finally, in order to enhance the performance and the discriminative ability of the new MTS, the convolutional structure and fully connected module are established. Furthermore, to better validate the effectiveness of MTS forecasting, extensive experiments are executed on two different types of MTS such as stock data and shared bicycle data, respectively. The experimental results adequately demonstrate the effectiveness and the feasibility of the proposed method.

BiLSTM-MLAM: A Multi-Scale Time Series Prediction Model Based on Bi-LSTM and Local Attention Mechanism

“…Min Han et al [26] introduced an innovative method combining a hybrid variable selection algorithm with an enhanced Extreme Learning Machine (ELM) for predicting multivariate chaotic time series. In domains such as renewable energy and the power market, MLP and ELM demonstrate superior performance compared to traditional statistical approaches and machine learning methodologies, effectively processing intricate energy data, power demand, and market behavior to provide accurate predictions and decision support [27][28][29][30]. However, due to their simplistic structure, ANNs have specific limitations in time feature extraction.…”

Computer Simulation of Freeze-thaw Cycle in Long'en Hall of Zhaoling Mausoleum in Shenyang City, China

“…(Chen et al 2023) proposed TSMixer with all-MLP architecture to efficiently utilize cross-variate and auxiliary information to improve the performance of time series forecasting. LightTS (Zhang et al 2022) is dedicated to solving multivariate time series forecasting problems, and it can efficiently handle very long input series. (Yi et al 2023) explores MLP in the frequency domain for time series forecasting and proposes a novel architecture for FreTS that includes two phases: domain conversion and frequency learning.…”

Research on Stock Prediction Based on CED-PSO-StockNet Time Series Model

“…Huihui Z and other scholars have proposed a new framework based on Encoder-Decoder for complex and variable temporal sequences. In this structure, the encoding and decoding processes use gated recursive units (GRUs) as the main unit structure to extract useful continuous feature information [10] . He K et al proposed using a combination of Convolutional Neural Network (CNN) and Long Short Term Memory Network (LSTM) to predict stock prices, constructing a feature extraction layer using CNN to extract features, and inputting the extracted features into LSTM to obtain time series information [11] ; Eapen J and other scholars proposed several variants of multi pipeline and single pipeline deep learning models based on different CNN kernel sizes and GRU unit numbers [12] ; Ming Che Lee proposed a stock trend prediction model that incorporates both GRU and attention mechanisms, offering a comprehensive model designed to handle the complexities of financial markets [13] ; Wenjie LU and other scholars proposed a new combination model consisting of three parts, which uses convolutional neural networks (CNN) to collect factors that affect stock prices, attention mechanisms are used to calculate the impact of stock data on stock prices at different times, and Gated Loop Units (GRU) are used for stock price prediction [14] .…”