SPDNet: Seasonal-Periodic Decomposition Network for Advanced Residential Demand Forecasting

TL;DR

SPDNet tackles residential electricity demand forecasting by explicitly modeling multi-scale temporal patterns. It combines a Seasonal-Trend Decomposition Module (STDM) with a Periodical Decomposition Module (PDM) that uses FFT to identify dominant periods and process the data through parallel short-term, periodic, and long-term branches, with outputs fused as $\hat{X} = \alpha_1 \hat{X}_{\text{PDM}} + \alpha_2 \hat{X}_{\text{STDM}}$. The approach yields state-of-the-art accuracy and computational efficiency on real residential load data, validated against a broad set of baselines and across multiple horizons, sequence lengths, and loads. The work advances practical forecasting for smart grids by isolating and modeling seasonal and periodic components, and provides code for replication.

Abstract

Residential electricity demand forecasting is critical for efficient energy management and grid stability. Accurate predictions enable utility companies to optimize planning and operations. However, real-world residential electricity demand data often exhibit intricate temporal variability, including multiple seasonalities, periodicities, and abrupt fluctuations, which pose significant challenges for forecasting models. Previous models that rely on statistical methods, recurrent, convolutional neural networks, and transformers often struggle to capture these intricate temporal dynamics. To address these challenges, we propose the Seasonal-Periodic Decomposition Network (SPDNet), a novel deep learning framework consisting of two main modules. The first is the Seasonal-Trend Decomposition Module (STDM), which decomposes the input data into trend, seasonal, and residual components. The second is the Periodical Decomposition Module (PDM), which employs the Fast Fourier Transform to identify the dominant periods. For each dominant period, 1D input data is reshaped into a 2D tensor, where rows represent periods and columns correspond to frequencies. The 2D representations are then processed through three submodules: a 1D convolution to capture sharp fluctuations, a transformer-based encoder to model global patterns, and a 2D convolution to capture interactions between periods. Extensive experiments conducted on real-world residential electricity load data demonstrate that SPDNet outperforms traditional and advanced models in both forecasting accuracy and computational efficiency. The code is available in this repository: https://github.com/Tims2D/SPDNet.

Figures (5)

• Figure 1: Overview of the proposed Seasonal-Periodic Decomposition Network (SPDNet) for residential electricity demand forecasting.
• Figure 2: Overview of the Periodical Decomposition Module (PDM) Framework to reshape the 1D input tensor into $k$ distinct 2D tensors.
• Figure 3: Overall architecture of transformer encoder processing to capture global dependencies.
• Figure 4: Forecasted and actual electricity demand for Load 1 at sequence length $S=96$ and prediction horizon $P=96$. Legends are shown only in the first subplot for clarity.
• Figure 5: Average training time (seconds) per epoch for different models. All experiments are trained with an input length of 720, batch size of 32, and prediction horizons of {1,4,24,48,96,192,384,672,1344,2688}.