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

TL;DR

PowerMamba introduces a deep state-space framework for multivariate time series forecasting in electric power systems, achieving high accuracy with substantially fewer parameters than Transformer baselines. It integrates a decomposition module and an external-predictions processor, enabling effective use of high-resolution forecasts without inflating model size. The authors release GridSet, a five-year ERCOT dataset with 22 core series (and 262-feature extended versions) and an open benchmarking toolbox, facilitating reproducible evaluation across loads, prices, ancillary services, and renewables. Results show PowerMamba outperforms baselines across short- and long-horizon tasks and maintains robustness to external forecast noise, with strong generalization evidenced by PJM experiments.

Abstract

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%.

Figures (11)

• Figure 1: Average Mean Squared Error (MSE) comparison between PowerMamba and state-of-the-art baselines with a context length of 240 hours and a 24-hour prediction window. The circle center represents the maximum possible error, and closer to the boundary indicates better performance.
• Figure 2: PowerMamba architecture. This model uses decompositions along with linear projections, and a combination of normal and inverse Mamba blocks to improve prediction.
• Figure 3: The structure of the dataset with external predictions. For a context window ending at 4 PM, the corresponding 24-hour external predictions are extracted.
• Figure 4: MSE of all models for different prediction window sizes, with context length $L=240$ and no external predictions.
• Figure 5: The number of model parameters with and without external predictions, in log scale.