A Global Optimization Algorithm based on Interval Analysis for Training Feedforward Neural Networks
A new Feedforward neural networks (FNN) global optimization algorithm based on interval analysis was presented in this paper. In the course of neural networks training with BP algorithm, the error function may be stuck in a local minimum. In order to solve this problem, interval analysis was brought into FNN to search a global optimal result. A new interval extension model was proposed to create a more precise and narrower interval domain. And three discard methods were employed together to accelerate the rate of convergence in the FNN training. In the presented algorithm, the objective function gradient was utilized sufficiently in both interval extension and discard methods to reduce the computing time. Simulation results are given to show the feasibility and performance of new algorithm.
Hongru Li Hailong Li Yina Du
Key Laboratory of Integrated Automation of Process Industry (Northeastern University) Ministry of Ed Key Laboratory of Integrated Automation of Process Industry (Northeastern University) Ministry of Ed
国际会议
南宁
英文
2007-07-20（万方平台首次上网日期，不代表论文的发表时间）