A Domain Knowledge-Guided Industrial Large Model Framework: A Case Study in Battery Health Estimation and Recycling
Bingyang Chen, Haidong Shao, Yao Qin, Yang Jin, Xinming Hu
Abstract
Accurate prediction of battery state of health (SOH) is essential for optimizing recycling processes. However, existing deep learning models often struggle to adapt to batteries with diverse materials and operating conditions. Some studies have explored the use of large language models to improve model generalization, but there are still limitations in effectively embedding battery measurement data, incorporating degradation domain knowledge, and adapting model parameters to specific battery characteristics. Therefore, this article introduces a domain knowledge-guided industrial large model (DK-ILM) framework, consisting of sequence reprogramming embedding (SRE), domain knowledge embedding (DKE), and sparsity-aware adapter (SAA), to enhance the generalization capability of SOH prediction. The SRE jointly encodes battery discharge data and relaxation voltage to generate effective measurement embeddings. The DKE models battery degradation mechanisms as prompt information, building domain knowledge embeddings that guide the model in identifying key features of battery health. The SAA combines both embeddings and employs a sparse attention mechanism to focus on key information, improving the model’s adaptability to the target battery. Experimental results on five battery datasets demonstrate that DK-ILM achieves high accuracy and robust generalization in SOH prediction tasks.