Based on the recurrent neural network and fuzzy system, dynamic T-S recurrent fuzzy neural network (DTRFNN) is proposed. The DTRFNN adopts BP algorithm for net weight learning and uses improved BP algorithm to overcome the local minima. With dynamic system identification as an example, makes a simulation research and compares it with general fuzzy neural network. The result shows that identification error of DTRFNN is smaller than general fuzzy neural network and achieves better identification effect. When the DTRFNN applying to Soft metal temperature measurement, it well realizes the on-line detection.