Benchmarking State Space Models, Transformers, and Recurrent Networks for US Grid Forecasting

TL;DR

This paper benchmarks five neural network architectures, including state space models, Transformers, and LSTM, for US power grid forecasting, revealing that model effectiveness varies with data type and available inputs.

Contribution

It provides a comprehensive comparison of modern neural architectures for power grid forecasting, highlighting how data availability influences model performance and guiding model selection.

Findings

PatchTST and state space models excel with only historical load data.

Adding weather data reverses model rankings, favoring iTransformer.

Model suitability varies across different energy and price forecasting tasks.

Abstract

Selecting the right deep learning model for power grid forecasting is challenging, as performance heavily depends on the data available to the operator. This paper presents a comprehensive benchmark of five modern neural architectures: two state space models (PowerMamba, S-Mamba), two Transformers (iTransformer, PatchTST), and a traditional LSTM. We evaluate these models on hourly electricity demand across six diverse US power grids for forecast windows between 24 and 168 hours. To ensure a fair comparison, we adapt each model with specialized temporal processing and a modular layer that cleanly integrates weather covariates. Our results reveal that there is no single best model for all situations. When forecasting using only historical load, PatchTST and the state space models provide the highest accuracy. However, when explicit weather data is added to the inputs, the rankings reverse: iTransformer improves its accuracy three times more efficiently than PatchTST. By controlling for model size, we confirm that this advantage stems from the architecture's inherent ability to mix information across different variables. Extending our evaluation to solar generation, wind power, and wholesale prices further demonstrates that model rankings depend on the forecast task: PatchTST excels on highly rhythmic signals like solar, while state space models are better suited for the chaotic fluctuations of wind and price. Ultimately, this benchmark provides grid operators with actionable guidelines for selecting the optimal forecasting architecture based on their specific data environments.