Abstract: Time/space separation methods typically adopt linear decomposition and reconstruction models that rely heavily on spatial basis functions to decompose and reconstruct the time-space domain of thermal processes, which are unable to handle highly nonlinear thermal dynamics. In order to solve this problem, this paper proposes a nonlinear spatiotemporal modeling method combining t-distributed stochastic neighbor embedding (t-SNE) and a broad learning system (BLS). A parametric t-SNE is designed to convert the spatiotemporal domain of lithium-ion battery (LIB) thermal processes into a temporal domain. Compared with traditional linear decomposition models, t-SNE can better preserves the nonlinear characteristics of spatiotemporal temperature data. Then BLS is used to construct a low-order temporal model. Finally, to overcome the non-invertibility of the t-SNE dimensionality reduction, a BLS-based nonlinear reconstruction model is further developed to reconstruct the spatiotemporal domain. The sparrow search algorithm (SSA) is used to optimize the parameters of BLS, which improves the accuracy and generalization performance of the model. Experimental results on ternary LIBs show that the root-mean-square error obtained by the proposed method decreases by 6.54% and the computation time decreases by 29.95% compared with traditional spatiotemporal modeling methods. These results verify the superiority and effectiveness of the proposed method for modeling nonlinear thermal processes in LIBs.