TSI: A Multi-View Representation Learning Approach for Time Series Forecasting

TL;DR

This work tackles long-horizon time series forecasting by learning a multi-view representation that blends trend-seasonality (TS) information with nonlinear independent components via ICA. The proposed TSI framework builds a unified latent representation H = [H_tr; H_s; H_i], combining a TCN-based trend encoder, a learnable Fourier seasonal encoder, and a VAE/FastICA-based independent component extractor, reinforced by a MoCo-inspired contrastive objective. Empirical results across six real-world datasets show robust improvements in MSE and MAE over state-of-the-art baselines, particularly at long horizons, demonstrating the method's effectiveness and scalability. By bridging TS decomposition with ICA, the work advances time series representation learning and points to future directions in causal analysis and ICA-driven forecasting.

Abstract

As the growing demand for long sequence time-series forecasting in real-world applications, such as electricity consumption planning, the significance of time series forecasting becomes increasingly crucial across various domains. This is highlighted by recent advancements in representation learning within the field. This study introduces a novel multi-view approach for time series forecasting that innovatively integrates trend and seasonal representations with an Independent Component Analysis (ICA)-based representation. Recognizing the limitations of existing methods in representing complex and high-dimensional time series data, this research addresses the challenge by combining TS (trend and seasonality) and ICA (independent components) perspectives. This approach offers a holistic understanding of time series data, going beyond traditional models that often miss nuanced, nonlinear relationships. The efficacy of TSI model is demonstrated through comprehensive testing on various benchmark datasets, where it shows superior performance over current state-of-the-art models, particularly in multivariate forecasting. This method not only enhances the accuracy of forecasting but also contributes significantly to the field by providing a more in-depth understanding of time series data. The research which uses ICA for a view lays the groundwork for further exploration and methodological advancements in time series forecasting, opening new avenues for research and practical applications.

Figures (3)

• Figure 1: Schematic of TSI, the proposed multi-view approach for time series forecasting
• Figure 2: The proposed multi-view time series forecasting model, incorporating Trend, Seasonal, and ICA encoders. The model's objective is to obtain a forecasted time series $Y^*$ that has the smallest difference from the test time series $Y$.
• Figure 3: Architectural Overview of the Time Series Decomposition Model. The left block extracts the trend component $H_{tr}$ using a Temporal Convolutional Network (TCN) and pooling. The middle block captures the seasonal component $H_{s}$ via FFT and a complex linear layer to encode frequency and phase. The right block extracts the independent component $H_{i}$ using a fully connected network with activations designed to minimize reconstruction error. This model decomposes time series into distinct features for robust representation.