Abstract: A dynamic T-S (Takgi-Sugeno) fuzzy Elman network (DTSFEN), which has the advantages of the T-S fuzzy model and Elman network, is proposed. This DTSFEN can process dynamic information with a globally convergent recursive structure. Moreover, an adaptive error back-propagation learning algorithm is designed to update the structure parameters and the fuzzy rule parameters, which can improve the learning efficiency of the DTSFEN. Then, the global convergence of the DTSFEN is proven using the Lyapunov stability theorem. Finally, the DTSFEN is used for nonlinear function approximation and sludge volume index (SVI) soft sensor. The experimental results show that the DTSFEN has a faster convergence rate and better accuracy and robustness than the orthogonal least squares (OLS) and Elman network.