基于Tan-Sigmoid函数参数调整的BP神经网络改进算法

Improved Algorithm of BP Neural Networks Based on Parameters Adjustment of Tan-Sigmoid Transfer Function

为提高BP神经网络的性能,对网络的联接权值W和神经元的tan-sigmoid转换函数的参数T、θ进行调整,使信息分布存储于权值矩阵及转换函数中,比传统的算法具有更强的非线性映射能力.经严密的数学推导,给出了最终的改进算法公式和1个预测需求量的算例,结果表明,改进后的算法能有效地减少隐层节点数,且能加快收敛速度和提高收敛精度.

In order to improve the performance of BP neural networks, the network weight matrix W and the parameters T,θ of neuron＇s tan-sigmoid transfer function are adjusted. Information is stored in the weight matrix and the transfer functions dispersedly. The improved algorithm has stronger nonlinear mapping eapability than traditional algorithm. The final formula of the improved algorithm is presented after rigorous mathematical deducing. An example of demand forecast is presented. The result indicates that the improved algorithm can reduce the nodes of hidden layer, accelerate the convergence and improve the convergence precision effectively.