PowerMamba: A Deep State Space Model and Comprehensive Benchmark for Time Series Prediction in Electric Power Systems

The electricity sector is undergoing substantial transformations due to the rising electrification of demand, enhanced integration of renewable energy resources, and the emergence of new technologies. These changes are rendering the electric grid more volatile and unpredictable, making it difficult to maintain reliable operations. In order to address these issues, advanced time series prediction models are needed for closing the gap between the forecasted and actual grid outcomes. In this paper, we introduce a multivariate time series prediction model that combines traditional state space models with deep learning methods to simultaneously capture and predict the underlying dynamics of multiple time series. Additionally, we design a time series processing module that incorporates high-resolution external forecasts into sequence-to-sequence prediction models, achieving this with negligible increases in size and no loss of accuracy. We also release an extended dataset spanning five years of load, electricity price, ancillary service price, and renewable generation. To complement this dataset, we provide an open-access toolbox that includes our proposed model, the dataset itself, and several state-of-the-art prediction models, thereby creating a unified framework for benchmarking advanced machine learning approaches. Our findings indicate that the proposed model outperforms existing models across various prediction tasks, improving state-of-the-art prediction error by an average of 7% and decreasing model parameters by 43%.

💡 Research Summary

The electric power sector is undergoing rapid transformation driven by demand electrification, high renewable penetration, and emerging technologies, which together increase volatility and make reliable operation more challenging. Traditional forecasting approaches that treat load, price, and renewable generation as separate problems are increasingly inadequate because they ignore the strong interdependencies among these variables. In response, the authors present PowerMamba, a multivariate time‑series forecasting framework that combines the efficiency of deep state‑space models (SSMs) with architectural innovations tailored to power‑system dynamics.

PowerMamba’s core consists of two parallel Mamba blocks – one standard and one “inverse” – that process the input sequence in two orthogonal ways. The forward block captures intra‑series temporal patterns within each variable, while the inverse block (operating on the transposed sequence) learns inter‑series relationships across time steps. This dual‑path design enables the model to simultaneously model the differential‑equation‑like dynamics of individual series (e.g., load following diurnal cycles) and the cross‑channel feedback loops that arise from market price‑response mechanisms.

A second major contribution is a time‑series processing module that seamlessly incorporates high‑resolution external forecasts (e.g., ERCOT day‑ahead load and renewable generation predictions). Rather than down‑sampling or inflating the feature dimension, the module decomposes each series into trend and seasonal components, projects them into a fixed‑size latent space, and injects the external forecasts at the same temporal granularity as the historical data. This preserves the model’s parameter count regardless of context length while still exploiting forward‑looking information that is especially valuable for highly stochastic series such as wind and solar.

To support rigorous evaluation, the authors release a comprehensive ERCOT‑based dataset covering five years of hourly data. The dataset includes 22 core series (zonal loads, electricity prices, ancillary service prices, wind and solar generation) and an extended version with 262 channels that embed external forecasts. Compared with existing public datasets, this collection offers longer temporal coverage, finer spatial granularity, and synchronized multivariate streams, making it well‑suited for modern deep‑learning models that thrive on large, diverse inputs.

Experimental results demonstrate that PowerMamba consistently outperforms a suite of strong baselines—including PatchTST, TimeMachine, iTransformer, and other transformer‑based models—across short‑ and medium‑term horizons. Using a context window of 240 hours and a 24‑hour prediction horizon, PowerMamba reduces average mean‑squared error by roughly 7 % while cutting the total number of trainable parameters by 43 % relative to the best existing Mamba‑based model and by 78 % relative to leading transformer architectures. Notably, the performance boost persists even for series that lack direct external forecasts, indicating that the processing module conveys useful indirect signals.

The authors also provide an open‑access toolbox that bundles the PowerMamba implementation, data loaders, evaluation scripts, and reference implementations of competing models. The codebase is publicly available on GitHub (https://github.com/alimenati/PowerMamba) with Docker support and detailed documentation, facilitating reproducibility and future extensions. Potential extensions discussed include incorporation of battery storage, demand‑response resources, and adaptation to distribution‑level datasets or other regional markets.

In summary, PowerMamba delivers a unified, efficient, and scalable solution for multivariate power‑system forecasting. By leveraging linear‑complexity SSMs, a dual‑path architecture, and high‑resolution external forecasts, it achieves superior accuracy with a markedly smaller footprint. The accompanying dataset and toolbox create a standardized benchmark that can accelerate research and deployment of advanced forecasting tools across the evolving electric grid.