Fourier-Enhanced Recurrent Neural Networks for Electrical Load Time Series Downscaling

TL;DR

This paper introduces a Fourier-enhanced RNN model that improves electrical load time series downscaling by integrating Fourier seasonal embeddings and self-attention, outperforming classical methods across multiple territories.

Contribution

The novel model combines Fourier embeddings and self-attention within an RNN for enhanced load downscaling accuracy, surpassing existing baselines.

Findings

Lower and flatter RMSE across horizons compared to Prophet.

Outperforms RNN variants without Fourier or attention.

Effective across multiple PJM territories.

Abstract

We present a Fourier-enhanced recurrent neural network (RNN) for downscaling electrical loads. The model combines (i) a recurrent backbone driven by low-resolution inputs, (ii) explicit Fourier seasonal embeddings fused in latent space, and (iii) a self-attention layer that captures dependencies among high-resolution components within each period. Across four PJM territories, the approach yields RMSE lower and flatter horizon-wise than classical Prophet baselines (with and without seasonality/LAA) and than RNN ablations without attention or Fourier features.