Short-Term Electricity-Load Forecasting by Deep Learning: A Comprehensive Survey
🤖 AI Summary
Short-term electric load forecasting (STELF) is highly time-varying and nonlinear due to external factors such as weather and consumer behavior, posing significant modeling challenges. This paper presents a systematic review of deep learning applications in STELF and proposes, for the first time, a comprehensive, full-stack classification framework covering temporal modeling, multi-source data fusion, uncertainty quantification, and edge deployment. We unify state-of-the-art architectures—including LSTM, GRU, TCN, Transformer, graph neural networks, and physics-informed hybrid models—while integrating data augmentation, missing-value imputation, and real-time inference optimization techniques. Based on an analysis of over 120 scholarly works, we establish a standardized evaluation benchmark and empirically demonstrate that lightweight models achieve an average MAPE below 1.8% in real-world grid deployments. The review identifies key bottlenecks and outlines future research directions, including model interpretability, cross-scenario generalizability, and energy-efficient edge deployment.
Application Category
Problem
Research questions and friction points this paper is trying to address.
Predicting short-term electricity demand accurately
Addressing non-linear load fluctuations from external factors
Surveying deep learning methods for load forecasting
Innovation
Methods, ideas, or system contributions that make the work stand out.
Deep learning for accurate electricity-load forecasting
Comprehensive data pre-processing and feature extraction
Optimized deep-learning modeling and evaluation
🔎 Similar Papers
No similar papers found.
