BP网络的改进研究

Research on improved BP network

针对标准BP神经网络收敛速度慢,学习精度不高的缺点,在标准BP神经网络算法中附加动量项,并以附加动量项的BP网络算法为基础,提出动量—自适应速率法,动量—可调激活函数法以及动量—自适应速率—激活函数法四种改进算法。以太阳黑子预测为实例分析四种改进算法在BP神经网络迭代次数减少,精度提高两方面的实际效果。事实证明,动量—可调激活函数算法对BP网络结构优化,提高收敛速度有明显效果。

To overcome the shortcomings of the standard BP network, this paper proposes four improved methods based on adding up inertia item, which are adaptive study rate method, variant sigmoid function method and adaptive study rate with variant sigmoid function method. The methods are verified by a typically real example which is the problem of sunspots forecasting from two aspects that are convergence speed and the precision. It is proved that variant sigmoid function method is the best one.