基于细胞集合的隐层冗余BP神经网络及其性能研究

Study on BPNN with redundant hidden neurons based on cell assembly and its performance

常规BP神经网络性能在很大程度上受制于其隐含层结构,本文对隐含层节点数的优化设置进行研究。利用大脑中基于细胞集合的信息处理机制,构建出一种隐含层具有冗余节点的BP神经网络(冗余BP网络),研究了其中隐含层冗余节点的设置方法,并在标准BP算法(SBP)基础上设计了一种批量式冗余BP算法(BBP)。结合模拟电路的输出特性与故障诊断,对所提网络的性能进行了实验测试,结果证实,所提冗余BP网络与常规BP网络相比,具有更好的逼近精度,更快的收敛速度和良好的数据泛化能力。

The performance of conventional back propagation neural network （BPNN） mainly depends on its hidden layer. In this paper, we study a simple approach to optimize the neurons in the hidden layer of BPNN. Utilizing the brain information processing mechanism based on cell assembly, a new BPNN with redundant hidden neurons was built. We first focused on how to settle the redundant neurons in the hidden layer, and then on how to train it. A batch B Palgorithm （BBP） was designed for training, which is different from standard B Palgorithm （SBP）. Finally, with the input-output relationshi Pand the fault diagnosis of a given analog circuit, the proposed BPNN was tested, and it is found that the proposed BPNN has better performance, such as better learning efficiency, more rapid convergence speed and stronger generalization ability.