Towards Continuous Power Forecasting: Practical Continual Learning for Real-World Energy Systems in Nonstationary Time Series

2026-06-23

AI Summary

This work addresses the persistent challenges in real-world power systems—namely, continuous data distribution shifts caused by weather variability, infrastructure upgrades, and evolving consumption patterns, compounded by limited historical data and the requirement for models to operate reliably over extended periods. For the first time, electric load forecasting is explicitly formulated as a continual learning problem. The authors propose an adaptive continual learning regression framework that enables models to self-adapt and accumulate knowledge in non-stationary environments without storing large volumes of historical data. Through systematic evaluation of six representative continual learning methods on real-world power datasets, the study demonstrates their effectiveness in mitigating catastrophic forgetting and maintaining long-term prediction stability, thereby advancing the practical deployment of continual learning in industrial-scale energy systems.

Abstract

Power forecasting models deployed in real-world energy markets must operate under nonstationary conditions, where data distributions continually evolve due to weather variability, infrastructure upgrades, and changing consumption behaviors. In practice, these models face strict operational constraints: historical data may be limited or unavailable for repeated retraining, and uninterrupted long-term service is often required. This paper addresses these challenges by proposing the paradigm of Continuous Power Forecasting, which views power forecasting as a continual learning problem rather than a static offline task. Based on an adaptive continual learning framework for regression, we systematically investigate the practical effectiveness of six representative continual learning approaches from three methodological categories. These approaches are evaluated under different realistic assumptions regarding data accessibility and update policies. Experimental validation on real-world power datasets demonstrates that continual learning enables forecasting models to self-adapt to distributional drift, accumulate knowledge over time, and mitigate catastrophic forgetting without relying on large-scale historical data storage. Beyond performance gains, our study provides practical insights into the stability and adaptation behaviors of different continual learning approaches under realistic operational constraints. Overall, this work illustrates how continual learning can be pragmatically integrated into industrial power forecasting pipelines, offering a scalable and sustainable solution for long-term deployment in dynamic environments.

Problem

• power forecasting
• nonstationary time series
• continual learning
• distributional drift
• catastrophic forgetting

Innovation

• Continual Learning
• Power Forecasting
• Nonstationary Time Series
• Catastrophic Forgetting
• Adaptive Regression