A dataset of Early fault diagnosis of electric vehicle batteries based on physics-informed DeepSeekV3 architecture
Description
Accurate early warning of battery failure is essential to ensure the safety of electric vehicles (EVs). However, the scarcity of early fault samples, the difficulty in capturing weak features, and the fact that pure data - driven models are prone to violate physical laws still pose challenges for early fault diagnosis. To address these issues, an unsupervised joint diagnosis framework combining generative pre-trained transformer and weighted one-class support vector machine (GPT-WOCSVM) is proposed for the first time in this paper. The charge data GPT (Charge-dataGPT) is constructed based on the physics-informed DeepSeekV3 architecture. It can use the charging data of the first 10 minutes to generate the charging sequence for the next 20 minutes in accordance with the electrochemical law through autoregression; WOCSVM is trained only with normal data, and early fault recognition is realized by judging whether the generated sequence deviates from the normal mode. This dataset contains the implementation code and model of GPT-WOCSVM.