基于BP神经网络隐层结构的研究及实例

Research and Example of Hidden Layer Structure Based on BP Neural Net-work

BP神经网络是神经网络中应用最为普遍的一种网络,随着人工智能技术的发展,各行各业也逐渐将BP神经网络运用在生活中,比如预测、推荐、识别等领域,都取得了一定的效果。但随着数据量的递增,BP神经网络也在进行预测时也有梯度下降等问题,许多专家也在不断对算法及网络结构进行调整。BP网络隐层结构的设计一直是不确定的,尤其是隐层单元数的确定缺乏理论依据,设计者大多依靠经验来确定。对于神经网络中BP网络的运用最为广泛,其中之一就是在函数收敛上的运用。文章主要是通过研究隐层层数和单元数的确定问题,来分析BP网络上的函数收敛性,通过比较在不同隐层层数和隐层节点下的收敛性来研究隐层结构对函数收敛性的影响,并将分析结果运用在股票预测中,实践表明,确定隐层节点数能在一定程度上改进预测误差。

BP neural network is the most commonly used network in neural networks.With the development of artificial intelligence technology,various industries have gradually applied BP ncural network in daily life,such as prediction,recommendation,recognition,and other fields,and have achieved certain results.But as the amount of data increases,the BP neural network also faces problems such as gradient descent when making predictions,and experts are constantly ad-justing the algorithm and nctwork structure.The design of the hidden layer structure in BP net-works has always been uncertain,espccially the determination of the number of hidden layer u-nits lacks theoretical basis,and designers mostly rely on experience to determine.The BP net-work is most widely used in neural networks,one of which is its application in function conver-gence.This article mainly analyzes the convergence of functions on BP networks by studying the problem of determining the number of hidden layers and units.By comparing the convergence under different hidden layer layers and hidden layer nodes,the influence of hidden layer struc-ture on function convergence is studied,and the analysis results are applied to stock price predic-tion.Practice has shown that determining the number of hidden layer nodes can improve predic-tion crrors to a certain extent.