A lithium-ion battery thermal process modeling framework based on evolutionary-designed relevance vector machine

With the rapid development of lithium-ion batteries (LIBs) in the renewable energy sector, distributed thermal process modeling has become a critical research focus. However, existing modeling methods primarily address challenges posed by the intrinsic nature of thermal processes while neglecting the noise issue inherent in the modeling procedure. To overcome this limitation, this paper proposes an evolutionary-designed relevance vector machine (RVM)-based thermal process modeling framework for LIBs. First, Karhunen–Loève (KL) decomposition is employed for time/space separation to extract the dominant spatial basis functions (SBFs) and their corresponding low-dimensional temporal coefficients. Subsequently, to mitigate noise interference, a dynamic temporal prediction model is constructed using RVM with denoising capability, which not only provides point predictions but also offers uncertainty quantification of the prediction results. Furthermore, a trade-off differential evolution algorithm is designed to optimize RVM hyperparameters, thereby improving prediction accuracy. Finally, an accurate temperature field distribution model for LIBs is obtained through time/space synthesis. In this manner, the proposed framework achieves accurate prediction of battery temperature distribution under noise interference. Experimental validation on a 3.5Ah ternary LIB confirms the effectiveness of the proposed framework.