一种改进的多层感知器神经网络技术

A modified multi-layer perceptron neural network technique.

多层感知器 (multi-layerperceptronnetworks ,MLPN)是一具有多层神经元、前馈、误差反传结构的神经网络 ,它的学习和预测能力受多方面因素的影响。首先我们从理论证明和数值分析的角度研究了传输函数、神经元的数目、网络层数及网络误差的迭代方式等与MLPN学习和预测能力的关系 ,对常规的MLPN作了改进 ;然后结合一个理论模型分析的例子 ,讨论了改进的MLPN对非线性函数的学习能力 ;最后 ,以某地野外磁测数据的去噪为实例 ,将本文介绍的神经网络技术用于插值 ,从而达到去噪的目的。

Multi-layer perceptron network (MLPN) is a neural net work with multi-layer neurons, feedback, and error backpropagation structures. Its learning and prediction ability is affected by several factors. First, the auth ors make researches on the relation of the MLPN learning and prediction ability with the propagation function, the number of neurons, the number of network laye rs, the iteration mode of network error, etc. , and modify the conventional MLPN . Then the learning ability of the modified MLPN for linear function is discusse d by combining with an example of theoretical model. The neural network techniqu e introduced here is used in interpolation for noise removal of real field magne tic data. The theoretical and real application results show that the modified ML PN of using random global optimization method to update network weights has a go od learning and prediction ability.