神经网络实现函数逼近的影响因素分析

Analysis of the Effect of BP Neural Network on Function Approximation

摘要: 基于向后传播算法的多层网络又称BP网络，因其易于实现，是目前应用最广的一种神经网络。本文主要研究基于MATLAB的BP网络在函数逼近的应用，分析其函数逼近效果的影响因素。经本文研究表明，人工神经网络随隐层层数增加，函数逼近误差减小，随隐层单元数增加，函数逼近误差减小，误差减小速率逐渐减慢。随训练精度减小，函数逼近误差减小，减小速率逐渐减慢，同时收敛步数增加。

Abstract: The multi-layer network based on backward propagation algorithm is also called BP network. Be-cause it is easy to implement, it is the most widely used neural network. This paper mainly studies the application of BP network based on MATLAB in function approximation, and analyzes the in-fluencing factors of its function approximation effect. The research results show that the artificial neural network increases with the number of hidden layers, and the function approximation error decreases. As the number of hidden layer elements increases, the function approximation error decreases, and the error reduction rate gradually decreases. As the training accuracy decreases, the function approximation error decreases, the decrease rate gradually decreases, and the number of convergence steps increases.