A multi-scale lithium-ion battery capacity prediction using mixture of experts and patch-based MLP

TL;DR

This work tackles lithium-ion battery capacity prediction under regeneration phenomena by introducing MSPMLP, a multi-scale model that combines a mixture of experts (MoE) with patch-based MLP blocks to capture both long-term degradation and local fluctuations. The MoE gate adaptively selects among patch-based MLPs with different patch sizes and uses top-$k$ activation to balance accuracy and efficiency, while intra-patch and inter-patch MLPs model local and global temporal patterns, respectively. On the NASA battery dataset using LOOCV, MSPMLP achieves a mean absolute error of $0.0078$, a 41.8% improvement over the best baseline, demonstrating strong generalization across varying operating conditions and regeneration events. The study highlights that multi-scale feature integration and selective expert activation are key for accurate, efficient early capacity prediction in battery health management, with potential for real-time deployment on real-world data.

Abstract

Lithium-ion battery health management has become increasingly important as the application of batteries expands. Precise forecasting of capacity degradation is critical for ensuring the healthy usage of batteries. In this paper, we innovatively propose MSPMLP, a multi-scale capacity prediction model utilizing the mixture of experts (MoE) architecture and patch-based multi-layer perceptron (MLP) blocks, to capture both the long-term degradation trend and local capacity regeneration phenomena. Specifically, we utilize patch-based MLP blocks with varying patch sizes to extract multi-scale features from the capacity sequence. Leveraging the MoE architecture, the model adaptively integrates the extracted features, thereby enhancing its capacity and expressiveness. Finally, the future battery capacity is predicted based on the integrated features, achieving high prediction accuracy and generalization. Experimental results on the public NASA dataset indicate that MSPMLP achieves a mean absolute error (MAE) of 0.0078, improving by 41.8\% compared to existing methods. These findings highlight that MSPMLP, owing to its multi-scale modeling capability and generalizability, provides a promising solution to the battery capacity prediction challenges caused by capacity regeneration phenomena and complex usage conditions. The code of this work is provided at https://github.com/LeiYuzhu/CapacityPredict.

Figures (13)

• Figure 1: MSPMLP overview.
• Figure 2: Multi-scale layer overview.
• Figure 3: Patch-based MLP block overview.
• Figure 4: The capacity degradation of NASA dataset.
• Figure 5: Train loss and test loss of B0005.