Hybrid Optimization Strategy for Radial Basis Function Neural Network Structure

WANG Ling,ZHENG Dazhong
DOI: https://doi.org/10.3321/j.issn:1000-0054.1999.07.013
1999-01-01
Abstract:To solve the structure optimization of a class of radial basis function neural netowrk, an efficient hybrid strategy is presented. First, the original problem is changed into a class of combinatorial optimization problem. Then, the hybrid strategy, which combines the parallel searching ability of genetic algorithm and the probabilistic jumping property of simulated annealing to improve the optimization properties and avoid trapping into local minima, selects good structure among structure space and calculates the weights using gradient method. Moreover, the constructing and deleting operators are introduced to make more flexibly searching process. In order to get better properties of generalization and anti noise, training error and testing error are considered together during the structure optimization process. Simulation results show that the hybrid strategy is very feasible and efficient.
What problem does this paper attempt to address?