Variable-step BP training algorithm with an adaptive momentum term

This paper proposes a new variable-step BP training algorithm which can overcome some disadvantages of the traditional BP training algorithm such as slow convergent speed and tendency of getting the local minima. In the new method, a nonlinear property term is introduced and its intensity coefficient is constructed as an adaptive nonlinear function, which processes the temperature ascending and temperature descending strategy. The simulation results show the efficiency of the new method.