[论文解读] Short-term electricity load forecasting with multi-frequency reconstruction diffusion

Qi Dong, Rubing Huang+4 more|arXiv (Cornell University)|Jan 10, 2026

Energy Load and Power Forecasting被引用 0

一句话总结

本论文提出基于多频重构的扩散（MFRD）模型用于短期电力负荷预测（STELF），将变分模态分解与扩散去噪框架结合，使用 Transformer/LSTM 去噪器，在 AEMO 与 ISO-NE 数据集上展示出更高的准确性。

ABSTRACT

Diffusion models have emerged as a powerful method in various applications. However, their application to Short-Term Electricity Load Forecasting (STELF) -- a typical scenario in energy systems -- remains largely unexplored. Considering the nonlinear and fluctuating characteristics of the load data, effectively utilizing the powerful modeling capabilities of diffusion models to enhance STELF accuracy remains a challenge. This paper proposes a novel diffusion model with multi-frequency reconstruction for STELF, referred to as the Multi-Frequency-Reconstruction-based Diffusion (MFRD) model. The MFRD model achieves accurate load forecasting through four key steps: (1) The original data is combined with the decomposed multi-frequency modes to form a new data representation; (2) The diffusion model adds noise to the new data, effectively reducing and weakening the noise in the original data; (3) The reverse process adopts a denoising network that combines Long Short-Term Memory (LSTM) and Transformer to enhance noise removal; and (4) The inference process generates the final predictions based on the trained denoising network. To validate the effectiveness of the MFRD model, we conducted experiments on two data platforms: Australian Energy Market Operator (AEMO) and Independent System Operator of New England (ISO-NE). The experimental results show that our model consistently outperforms the compared models.

研究动机与目标

• Develop a lightweight STELF model that relies only on load data without external features.
• Leverage multi-frequency decomposition to enrich input representations for improved forecasting.
• Propose a diffusion-based denoising framework with a Transformer and residual LSTM to improve robustness to noise.
• Validate MFRD on real-world datasets (AEMO and ISO-NE) and compare against multiple baselines.

提出的方法

• Create multi-frequency feature representations by applying Variational Mode Decomposition (VMD) to the load data and concatenating the resulting IMFs with the original signal.
• Apply a forward diffusion process to progressively add Gaussian noise to the augmented data, preparing it for denoising in the reverse process.
• Train a denoising network combining residual LSTM and Transformer architectures, with positional encoding and a Fourier-domain loss to capture frequency characteristics.
• Perform inference by iteratively denoising from Gaussian noise to reconstruct the prediction for the desired horizon.

Figure 1 : Overall MFRD framework.

实验结果

研究问题

• RQ1 Can a diffusion-based framework with multi-frequency features improve STELF accuracy without external inputs?
• RQ2 Does integrating VMD-derived frequency components with a diffusion-denoising network enhance robustness to noise in load data?
• RQ3 How does MFRD compare to traditional ML/DL methods for STELF on diverse datasets?

主要发现

• MFRD consistently outperforms baselines on the AEMO and ISO-NE datasets across evaluated metrics.
• Using multiple VMD modes (k values) significantly affects forecasting performance, with best results varying by region.
• The NSW dataset achieves strong performance with k = 4 in the reported table, indicating a regional sensitivity to frequency decomposition parameters.
• The model architecture leverages both global pattern capture (Transformer) and temporal dependency modeling (LSTM), enhancing denoising effectiveness.

a