Fault diagnosis of batteries in energy storage systems based on model fusion

With the widespread application of energy storage systems, thermal runaway of lithium-ion batteries has become an increasingly serious concern. Currently, most studies related to battery fault diagnosis focus on exploring external characteristics of the batteries to detect potential faults. However, some external characteristics like voltage inconsistency may not be strongly correlated with battery faults, thus traditional fault diagnosis fail to establish sufficient and necessary relationships with faults. To solve this problem, we propose a novel solution to the deficiencies of traditional battery fault diagnostics by considering both the internal states of batteries and risky usage behaviors throughout operational life. To fulfill this idea, an electrochemical model is established to capture battery internal states and a risk accumulation coefficient to characterize battery risky behaviors. On top of these, a self-attention mechanism model is employed to identify potential faults by fusing results from both aspects. The proposed method is trained and verified using data from real-world batteries and achieves an accuracy rate of 97.37%. Compared to existing methods, the proposed method offers higher fault identification accuracy with deeper understanding of fault development and triggering mechanisms.