Multivariate time-series (MTS) forecasting is a challenging task in many
real-world non-stationary dynamic scenarios. In addition to intra-series
temporal signals, the inter-series dependency also plays a crucial role in
shaping future trends. How to enable the model’s awareness of dependency
information has raised substantial research attention. Previous approaches have
either presupposed dependency constraints based on domain knowledge or imposed
them using real-time feature similarity. However, MTS data often exhibit both
enduring long-term static relationships and transient short-term interactions,
which mutually influence their evolving states. It is necessary to recognize
and incorporate the complementary dependencies for more accurate MTS
prediction. The frequency information in time series reflects the evolutionary
rules behind complex temporal dynamics, and different frequency components can
be used to well construct long-term and short-term interactive dependency
structures between variables. To this end, we propose FCDNet, a concise yet
effective framework for multivariate time-series forecasting. Specifically,
FCDNet overcomes the above limitations by applying two light-weight dependency
constructors to help extract long- and short-term dependency information
adaptively from multi-level frequency patterns. With the growth of input
variables, the number of trainable parameters in FCDNet only increases
linearly, which is conducive to the model’s scalability and avoids
over-fitting. Additionally, adopting a frequency-based perspective can
effectively mitigate the influence of noise within MTS data, which helps
capture more genuine dependencies. The experimental results on six real-world
datasets from multiple fields show that FCDNet significantly exceeds strong
baselines, with an average improvement of 6.82% on MAE, 4.98% on RMSE, and
4.91% on MAPE.
Multivariate time-series (MTS) forecasting is a complex task that requires considering both intra-series temporal signals and inter-series dependencies. The challenge lies in how to enable the model to understand and incorporate these dependency relationships. Previous approaches have either relied on domain knowledge or real-time feature similarity to impose dependency constraints. However, MTS data often exhibit both enduring long-term static relationships and transient short-term interactions, which mutually influence their evolving states.
The proposed FCDNet framework addresses these limitations by utilizing frequency information in time series data. By analyzing different frequency components, FCDNet is able to construct both long-term and short-term interactive dependency structures between variables. This approach allows for the adaptive extraction of dependency information from multi-level frequency patterns, improving the accuracy of MTS prediction.
One of the strengths of FCDNet is its scalability. As the number of input variables increases, the number of trainable parameters in FCDNet only increases linearly. This makes the model more scalable and helps prevent overfitting, a common issue in complex forecasting models.
Furthermore, FCDNet adopts a frequency-based perspective, which proves effective in mitigating the influence of noise within MTS data. By focusing on genuine dependencies, the model is able to capture more accurate and reliable patterns.
The experimental results on six real-world datasets from multiple fields demonstrate the effectiveness of FCDNet. It outperforms strong baseline models, achieving an average improvement of 6.82% on MAE (Mean Absolute Error), 4.98% on RMSE (Root Mean Square Error), and 4.91% on MAPE (Mean Absolute Percentage Error).
The multi-disciplinary nature of the concepts presented in this content is worth noting. MTS forecasting involves knowledge from fields such as time-series analysis, signal processing, and machine learning. By integrating these disciplines, FCDNet provides a comprehensive framework that leverages frequency information to improve the accuracy of multivariate time-series forecasting.