Title :
An Efficient Activation Function for BP Neural Network
Author :
Hu, Jie ; Zeng, Xiangjin
Author_Institution :
Sch. of Sci., Wuhan Univ. of Technol., Wuhan
Abstract :
In traditional BP algorithm, sigmoid activation function outputs are restricted in interval [0, 1], and BP algorithm converges slow and has very low accuracy near 0 and 1. A new sigmoid activation function is put forward and applied to the forecast of short-term traffic flow. Simulation shows that the proposed activation function has overcome the above two shortcomings and got better effect than the traditional one.
Keywords :
backpropagation; neural nets; traffic information systems; BP neural network; short-term traffic flow forecasting; sigmoid activation function; Accuracy; Algorithm design and analysis; Artificial neural networks; Convergence; Feedforward neural networks; Feedforward systems; Neural networks; Telecommunication traffic; Traffic control;
Conference_Titel :
Intelligent Systems and Applications, 2009. ISA 2009. International Workshop on
Conference_Location :
Wuhan
Print_ISBN :
978-1-4244-3893-8
Electronic_ISBN :
978-1-4244-3894-5
DOI :
10.1109/IWISA.2009.5072710
