Hybrid CNN-transformer for multivariate time series forecasting with static and dynamic factors

The growing prevalence of time series data highlights the need for forecasting models that can effectively capture from external influencing factors. Multivariate time series forecasting, however, often faces challenges as existing methods tend to overlook the critical role of static factors, limiting their ability to model the interplay between contextual and temporal dependencies. To address this limitation, this paper proposes a hybrid CNN-Transformer model that integrates static and dynamic factors, leveraging convolutional operations to extract localized features from static inputs and a Transformer-based architecture to capture long-term dependencies in dynamic data. Experiments demonstrate the model’s superior performance, achieving significant improvements in mean squared error (MSE) and mean absolute error (MAE) compared to baseline methods. Ablation studies further highlights the importance of incorporating static factors, revealing substantial performance degradation when these components are excluded. These findings establish the proposed model as a robust solution for complex multivariate time series forecasting tasks.