Adaptive Multi-Task Learning for Battery Health Prediction: A BiLSTM Framework with Bayesian Optimization and Dynamic Task Weighting

Abstract

Accurate and reliable prediction of lithium-ion battery health is critical for enhancing battery lifespan, ensuring operational safety, and optimizing energy management in electric vehicles and energy storage systems. However, battery degradation is inherently complex, characterized by nonlinear and battery-specific aging patterns, which pose significant challenges for traditional single-task learning approaches. In this study, we propose a Multi-Task Learning Bidirectional Long Short-Term Memory framework that simultaneously predicts multiple key battery health indicators, including cycle life, voltage decay rate, and temperature change rate. To enhance model efficiency and adaptability, we introduce an adaptive prediction window, optimized through Bayesian hyperparameter tuning, allowing dynamic adjustment of window length based on individual battery degradation characteristics. Additionally, a dynamic task-weighting mechanism is incorporated to allocate computational resources based on task complexity and predictive uncertainty. The proposed framework is validated on the NASA battery dataset, demonstrating superior predictive performance compared to conventional machine learning models. Experimental results show that the framework achieves significant improvements in predictive accuracy across all health indicators, with the adaptive prediction window and dynamic task weighting contributing to enhanced generalization and robustness. These findings underscore the potential of framework for real-time battery health monitoring and predictive maintenance.

Keywords: Multi-task learning, Bidirectional Long Short-Term Memory, Bayesian optimization, Adaptive sliding window, Battery health prediction