Abstract:Time series forecasting is crucial in many fields, yet current deep learning models struggle with noise, data sparsity, and capturing complex multi-scale patterns. This paper presents MFF-FTNet, a novel framework addressing these challenges by combining contrastive learning with multi-scale feature extraction across both frequency and time domains. MFF-FTNet introduces an adaptive noise augmentation strategy that adjusts scaling and shifting factors based on the statistical properties of the original time series data, enhancing model resilience to noise. The architecture is built around two complementary modules: a Frequency-Aware Contrastive Module (FACM) that refines spectral representations through frequency selection and contrastive learning, and a Complementary Time Domain Contrastive Module (CTCM) that captures both short- and long-term dependencies using multi-scale convolutions and feature fusion. A unified feature representation strategy enables robust contrastive learning across domains, creating an enriched framework for accurate forecasting. Extensive experiments on five real-world datasets demonstrate that MFF-FTNet significantly outperforms state-of-the-art models, achieving a 7.7% MSE improvement on multivariate tasks. These findings underscore MFF-FTNet's effectiveness in modeling complex temporal patterns and managing noise and sparsity, providing a comprehensive solution for both long- and short-term forecasting.

What problem does this paper attempt to address?

This paper attempts to solve several key problems in time - series prediction, specifically including: 1. **Insufficient noise robustness**: Many existing time - series prediction models perform poorly when dealing with noisy or non - stationary data. These models may over - smooth the data or be too sensitive to noise, thus affecting the prediction accuracy. 2. **Inadequate capture of multi - scale patterns**: Existing time - series prediction models often have difficulty effectively capturing multi - scale patterns, especially those covering both short - term and long - term dependencies simultaneously. Current methods may only focus on short - term fluctuations or long - term trends, but fail to effectively combine the two. 3. **Inability to effectively integrate time - domain and frequency - domain features**: Although some models focus on time - domain or frequency - domain features, few frameworks can integrate these two features in a complementary manner. This leads to limitations in the understanding of time - series data, as they cannot fully utilize the complementary advantages of the two domains. To solve the above problems, the paper proposes a new time - series prediction model named **MFF - FTNet**. MFF - FTNet improves the model's robustness and prediction accuracy by combining contrastive learning and multi - scale feature extraction to perform feature fusion in the frequency domain and time domain respectively. Specifically, MFF - FTNet introduces the following innovations: - **Adaptive noise enhancement strategy**: Dynamically adjust the scaling and translation factors according to the statistical properties of the original time - series data, enhancing the model's robustness to noise. - **Frequency - Aware Contrast Module (FACM)**: Improve the spectral representation through frequency selection and contrastive learning, especially apply mask operations in the frequency domain to filter out irrelevant high - frequency noise and generate more robust frequency - domain features. - **Complementary Time - Domain Contrast Module (CTCM)**: Utilize multi - scale convolution and feature fusion to capture short - range and long - range dependencies, and identify subtle but important differences between different time points and feature dimensions through contrastive learning, thereby improving the model's ability to learn complex time structures. - **Unified feature representation strategy**: Integrate time - domain and frequency - domain features, provide robust cross - domain contrastive learning, and construct a comprehensive and rich prediction framework, improving the model performance. Through these methods, MFF - FTNet has been extensively experimented on multiple real - world datasets, and the results show that it significantly outperforms the existing state - of - the - art models, especially achieving a 7.7% improvement in mean - squared error (MSE) in multivariate tasks. This proves the effectiveness of MFF - FTNet in modeling complex time patterns, managing noise and sparsity, and provides a comprehensive solution for long - term and short - term prediction.

MFF-FTNet: Multi-scale Feature Fusion across Frequency and Temporal Domains for Time Series Forecasting

TFEformer: Temporal Feature Enhanced Transformer for Multivariate Time Series Forecasting

MMFNet: Multi-Scale Frequency Masking Neural Network for Multivariate Time Series Forecasting

Adaptive Multi-Scale TF-net for High-Resolution Time-Frequency Representations

A time-frequency fusion model for multi-channel speech enhancement

Deep learning model based on multi-scale feature fusion for precipitation nowcasting

CTFNet: Long-Sequence Time-Series Forecasting Based on Convolution and Time–Frequency Analysis

Adaptive feature fusion for time series classification

MFF-Net: Towards Efficient Monocular Depth Completion With Multi-Modal Feature Fusion

MSGNet: Learning Multi-Scale Inter-Series Correlations for Multivariate Time Series Forecasting

MFFnet: A Seismic Phase Picking Network Based on Multiple Feature Fusion

Towards Better Forecasting by Fusing Near and Distant Future Visions

FL-Net: A multi-scale cross-decomposition network with frequency external attention for long-term time series forecasting

FDNet: Focal Decomposed Network for Efficient, Robust and Practical Time Series Forecasting

TFDNet: Time-Frequency Enhanced Decomposed Network for Long-term Time Series Forecasting

Adaptive Convolutional Forecasting Network Based on Time Series Feature-Driven

Masked Multi-Step Multivariate Time Series Forecasting with Future Information

Not All Frequencies Are Created Equal:Towards a Dynamic Fusion of Frequencies in Time-Series Forecasting

GA-FCFNN: A new forecasting method combining feature selection methods and feedforward neural networks using genetic algorithms

FCNet: Fully Complex Network for Time Series Forecasting

MultiWaveNet: A Long Time Series Forecasting Framework Based on Multi-Scale Analysis and Multi-Channel Feature Fusion