改进的粒子群动态过程神经网络及其应用

为克服前向过程神经网络收敛速度慢、精度低的问题,本文提出了一种基于改进的粒子群动态过程神经网络(IDPNN)。对于给定的全连接的过程神经网络,通过IPSO优化其连接权值和网络结构删除冗余连接使之成为部分连接的过程神经网络系统,从而降低了计算成本。将经过IPSO训练的动态过程神经网络应用于Iris模式分类问题,结果表明,改进的粒子群动态过程神经网络具有较高的收敛速度和精确性。

Modified particle swarm dynamic process neural network and its application

To overcome the shortcomings of slow convergent speed and low accuracy of the feedforward process neural network, an Improved Particle Swarm Optimization (IPSO) is proposed to train the dynamic process neural network. A fully connected feedforward dynamic process neural network can be converted into partially connected network by tuning the structure and choosing the connection weights of PNN simultaneously. This will result in significant cost reduction in implementation of the neural network. The dynamic process neural network trained by IPSO has been applied to Iris pattern classification. Results show that the modified particle swarm dynamic process neural network can accelerate the convergent speed and improve the accuracy.