Towards Sustainable Power Systems: Exploring the Opportunities of Multi-task Learning for Battery Degradation Forecasting
Abstract
The task of developing a reliable and long-lasting battery system remains a major obstacle to the expansion of the electric vehicle fleet and the improvement of storage systems in renewable energy plants. To overcome this challenge, improving battery ageing models to accurately forecast the battery degradation trajectory is a complex but critical step. In this work, we introduce a novel sequence-to-sequence multi-task learning (MTL) method. Empirical results on battery ageing datasets show that our model enhances the data efficiency and precision of capacity and power degradation forecasts for lithium-ion battery cells. Furthermore, our model generates new insights into the trade-offs and key decisions for incorporating battery ageing models into the MTL paradigm.