To solve the tracking control problem for velocity curve of urban rail vehicle, an improved predictive function control algorithm IPFC is proposed. Step function and Morlet function are selected as two base functions. According to the change degree of target velocity curve, a strategy for basis function selection of predictive function control is given, which can switch wavelet and step basis function. In addition, an adaptive nonlinear online softening factor adjustment strategy based on the fuzzy satisfaction of system performance and optimization factor is proposed. This strategy can further improve the tracking control performance by using optimization factor. Taking the instance of velocity curve tracking control from Lvshun New Port to Tieshan Town pertain of the urban rail transit line No. 12 in Dalian as the test object, the hardware-in-the-loop test results show that the proposed IPFC can improve control performance of control system significantly. The quality indexes, such as energy conservation, accurate parking, punctuality and comfort, have obvious improvement effectiveness, especially for punctual and precise parking. Compared with the traditional improved algorithms widely used in practice, they are improved to be more than 55% .