The field of time series forecasting is experiencing significant advancements, driven by the development of innovative machine learning models and techniques. Researchers are focusing on improving forecast accuracy, stability, and interpretability, while also addressing challenges such as non-stationarity, heteroskedasticity, and concept drift. Noteworthy papers in this area include: N-BEATS-MOE, which introduces a Mixture-of-Experts layer to improve forecasting performance on heterogeneous time series. TriForecaster, a framework that leverages a Mixture of Experts approach to address regional, contextual, and temporal variations in multi-region electric load forecasting. Wavelet Mixture of Experts for Time Series Forecasting, which combines wavelet transforms with a Mixture of Experts framework to capture periodic and non-stationary characteristics of data. These advancements have the potential to improve the accuracy and reliability of time series forecasting models, with applications in various fields such as finance, energy, and transportation.
Advancements in Time Series Forecasting
Sources
Dual Signal Decomposition of Stochastic Time Series
N-BEATS-MOE: N-BEATS with a Mixture-of-Experts Layer for Heterogeneous Time Series Forecasting
Adaptive Fine-Tuning via Pattern Specialization for Deep Time Series Forecasting
Multi-grained spatial-temporal feature complementarity for accurate online cellular traffic prediction
Wavelet Mixture of Experts for Time Series Forecasting
Stationarity Exploration for Multivariate Time Series Forecasting
HiSTM: Hierarchical Spatiotemporal Mamba for Cellular Traffic Forecasting
TriForecaster: A Mixture of Experts Framework for Multi-Region Electric Load Forecasting with Tri-dimensional Specialization
Measuring Time Series Forecast Stability for Demand Planning
Predictive Position Control for Movable Antenna Arrays in UAV Communications: A Spatio-Temporal Transformer-LSTM Framework