A Brain-Inspired Spiking Network Framework Based on Multi-Time-Step Self-Attention for Lithium-Ion Batteries Capacity Prediction

• Published in: IEEE transactions on consumer electronics Vol. 70; no. 1; pp. 3008 - 3017
• Main Authors: Wang, Huan, Sun, Muxia, Li, Yan-Fu
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.02.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Artificial neural networks; battery health monitoring; Biological system modeling; Bionics; Brain; Brain modeling; Computational modeling; Convolution; Encoding; Energy consumption; Lithium-ion batteries; Lithium-ion battery; Neural networks; Neurons; Power consumption; Rechargeable batteries; self-attention; Spiking; spiking neural network
• ISSN: 0098-3063; 1558-4127
• DOI: 10.1109/TCE.2023.3279135

Lithium-ion (Li-ion) batteries have gained widespread usage in numerous consumer electronic products and have significantly contributed to the growth of related industries. Due to the instability issues which might cause explosion or fire, it is critical to ensure the safety and reliability of Li-ion batteries via health monitoring. While artificial neural networks (ANNs) have proven successful in battery health monitoring, they suffer from drawbacks such as high energy consumption and poor generalization. Alternatively, a recent well-developed highly bionic model, i.e., brain-inspired spiking neural networks (SNN), has an excellent simulation of the spatiotemporal feature learning abilities and low power consumption characteristics of biological brains. In this study, we propose a multi-time-step self-attention spiking network framework (MSSA-SNN) for battery monitoring. In particular, the SNN architecture-based self-attention module enables a fully encoding of the global spiking features while optimizes the synaptic weights from a global perspective. This study conducts experiments on two coin Li-ion battery datasets. The results demonstrate that the brain-inspired MSSA-SNN can accurately detect battery degradation trends with an extremely low energy consumption rate, which makes it well-suited for energy-constrained consumer electronics.