Learning When Not to Use a Battery: Multihorizon Failure Intelligence
Lopullinen julkaistu versio - 1.87 MB
Shikdar, T. A. & Laaksonen, H. (2026). Learning When Not to Use a Battery: Multihorizon Failure Intelligence. International Transactions on Electrical Energy Systems. 2026(1). https://doi.org/10.1155/etep/6000810
© 2026 Tareq Anwar Shikdar and Hannu Laaksonen. International Transactions on Electrical Energy Systems published by John Wiley & Sons Ltd. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. https://creativecommons.org/licenses/by/4.0/
Lataukset8
Pysyvä osoite
Kuvaus
Battery energy storage systems (BESSs) are increasingly dispatched to provide flexibility services in renewable-dominated power systems. However, conventional battery analytics focuses on state-of-health (SOH) and remaining useful life (RUL), which describe long-term degradation rather than short-term operational safety. In practical grid operation, the relevant decision is whether a battery can reliably complete a specific service commitment within a predefined time horizon. This paper reformulates battery prognostics as an operational reliability estimation problem and proposes a decision-oriented framework based on multihorizon discrete-time hazard learning. The method predicts the probability of failure within a service window and applies probability calibration to obtain a trustworthy risk metric. The calibrated risk is integrated into a reliability-aware dispatch policy that regulates participation in flexibility according to an acceptable failure tolerance. The framework is evaluated on 37 lithium-ion batteries using leave-battery-out cross-validation to ensure cross-device generalization. The hazard model achieves strong discrimination performance, reaching an area under the ROC curve (AUC) of 0.944 before calibration for the 20-cycle horizon. After probability calibration, the operational probabilities used for decision-making achieve an AUC of 0.891 with a Brier score of 0.064, demonstrating reliable probabilistic risk estimation across service horizons. When incorporated into dispatch decisions, the reliability-aware policy reduces the operational failure rate from 10.3% to 2.95% and increases delivered energy from 43.4 to 47.0 kWh compared with non-reliability-aware operation. An early-life analysis further shows that operational risk becomes predictable only after degradation observability emerges, distinguishing reliability estimation from long-term lifetime prediction. The results demonstrate that battery management for power systems should transition from predicting lifetime to controlling operational risk. The proposed framework provides an interpretable reliability signal enabling safe participation of storage assets in flexibility markets and bridges the gap between battery diagnostics and operational decision-making.
Emojulkaisu
ISBN
ISSN
2050-7038
Aihealue
Kausijulkaisu
International transactions on electrical energy systems|2026
OKM-julkaisutyyppi
A1 Alkuperäisartikkeli tieteellisessä aikakauslehdessä (vertaisarvioitu)