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

Sunki Hong, Jisoo Lee, Yuanyuan Shi · 2026

Read Benchmarking State Space Models, Transformers, and Recurrent Networks for US Grid Forecasting by Sunki Hong, Jisoo Lee, Yuanyuan Shi (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-21415.txt.

Description

Sunki Hong and one co-author benchmark five modern neural architectures for power grid forecasting. The set holds two state space models, two transformers, and a classic LSTM. Choosing well is hard because the best model depends on the data the operator actually has. They test hourly electricity demand across six diverse United States grids for horizons of 24 to 168 hours.

No single model wins everywhere. With load history alone, PatchTST and the state space models lead. Adding explicit weather data flips the ranking. iTransformer then gains accuracy three times more efficiently than PatchTST. PatchTST also leads on rhythmic solar signals. State space models handle chaotic wind and price swings better. The benchmark thus gives grid operators practical rules for picking an architecture to match their data.

Catalog record

CREATORSunki Hong, Jisoo Lee, Yuanyuan Shi
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-21415

Browse the full library · About the Tower of Babel