Advances in Time Series Forecasting and Prognostics

The field of time series forecasting and prognostics is rapidly advancing, with a focus on developing innovative models and techniques to improve predictive accuracy and reliability. Recent research has emphasized the importance of capturing complex temporal dependencies, prioritizing critical features, and adapting to non-stationary conditions. Notable advancements include the integration of transformer-based architectures, graph neural networks, and frequency domain analysis to enhance forecasting performance. Furthermore, the development of novel frameworks and models, such as those incorporating contextual motifs, temporal stabilization, and frequency differencing, has shown significant promise in addressing the challenges of long-term forecasting and non-stationarity.

Noteworthy papers include: AWEMixer, which proposes an Adaptive Wavelet-Enhanced Mixer Network for long-term time series forecasting, demonstrating improved performance over state-of-the-art models. CometNet, a novel Contextual Motif-guided Long-term Time Series Forecasting framework, which effectively harnesses contextual information to strengthen long-term forecasting capability. Synapse, an arbitration framework for Time Series Foundational Models, which dynamically leverages a pool of models to construct a robust forecast distribution. DTAF, a dual-branch framework addressing non-stationarity in both temporal and frequency domains, generating robust forecasts that adapt to changing conditions.

Sources

Temporal convolutional and fusional transformer model with Bi-LSTM encoder-decoder for multi-time-window remaining useful life prediction

AWEMixer: Adaptive Wavelet-Enhanced Mixer Network for Long-Term Time Series Forecasting

Carbon Price Forecasting with Structural Breaks: A Comparative Study of Deep Learning Models

AWARE: Evaluating PriorityFresh Caching for Offline Emergency Warning Systems

Predicting and forecasting reactivity and flux using long short-term memory models in pebble bed reactors during run-in

No One-Model-Fits-All: Uncovering Spatio-Temporal Forecasting Trade-offs with Graph Neural Networks and Foundation Models

Synapse: Adaptive Arbitration of Complementary Expertise in Time Series Foundational Models

CometNet: Contextual Motif-guided Long-term Time Series Forecasting

Towards Non-Stationary Time Series Forecasting with Temporal Stabilization and Frequency Differencing

Learning-based Radio Link Failure Prediction Based on Measurement Dataset in Railway Environments

Spectral Predictability as a Fast Reliability Indicator for Time Series Forecasting Model Selection

An Improved Dual-Attention Transformer-LSTM for Small-Sample Prediction of Modal Frequency and Actual Anchor Radius in Micro Hemispherical Resonator Design

Context-Aware Management of IoT Nodes: Balancing Informational Value with Energy Usage

Built with on top of