径向基函数神经网络结构的混合优化策略

为了解决一类径向基函数神经网络的结构优化问题，提出了一种有效的混合优化策略。将结构优化问题转化为一类组合优化问题。利用结合遗传算法群体并行搜索能力和模拟退火概率突跳特性来改善优化效率并避免局部极小的混合策略。借助于有效的编码方式在结构解空间中优选网络结构。利用梯度下降计算隐层到输出层的连接权。增添和删除操作用于增加结构搜索的灵活性。为保证所得网络具有较好的推广能力，利用训练误差和检验误差的综合指标作为算法择取优良网络的依据。仿真研究表明，所提混合策略是快速有效的，且能保证网络具有较好的推广性和抗噪声能力。

Hybrid optimization strategy for radial basis function neural network structure

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.