径向基概率神经网络的结构优化算法研究

径向基概率神经网络（RBPNN）是在径向基函数神经网络（RBFNN）和概率神经网络（PNN）的基础上发展起来的一种新型的前馈神经网络（FNN）模型。该网络模型充分吸收了径向基函数神经网络和概率神经网络的优点,这种新的模型具有计算复杂度低、收敛速度快等优点。本文深入研究了径向基概率神经网络的结构优化算法,在遗传结构优化方法的基础上,提出一种新的两步学习算法,基于遗传算法的梯度学习算法。该算法一方面优化了网络结构,使网络结构尽可能的精简,另一方面有效地提高了网络的推广能力。

RESEARCH ON THE ALGORITHM FOR STRUCTURE OPTIMIZATION OF THE RADIAL BASIS PROBABILISTIC NEURAL NETWORKS

The radial basis probabilistic neural network （RBPNN） is a novel feed-forward neural network model developed from the radial basis function neural network （RBFNN） and the probabilistic neural network （PNN）. The RBPNN was proposed by absorbing the respective merits of the two latter neural network models, this new model is of the advantages such as lower computation complexity and faster convergent speed. The algorithms for the RBPNN＇ structure optimization were deeply studied, and the structure optimization theory for the RBPNNs was enriched. A novel two-step leaning algorithm combing genetic optimization with gradient-based algorithm is proposed. The designed network is not only parsimonious but also has better generalization performance.