正态分布在基于BP人工神经网络识别火山岩岩性中的应用

摘要

基于测井资料的BP神经网络法识别火山岩岩性应用比较广泛.若网络训练样本本身存在问题,将导致网络不易收敛、精度低.合理确定训练样本的输入值和期望输出值尤为重要.通过对大量的火山岩测井数据进行处理分析,发现部分测井响应特征参数在不同测段内基本上服从正态分布规律,据此应用正态分布理论,给出了合理确定训练样本方法和计算公式.计算结果表明,利用该方法进行火山岩岩性识别是可行的,该方法对其他相关领域也具有参考价值.%The method, based on nerve network to lithologic of volcanic rock lithology, was used widely. The precision ration of lithologic recognition was determined by the characters of the BP neural If there were many problems in training specimen itself, it would result in no-contraction and low accuracy of the network. It was very important for rational input and output values of training specimens. It was found out that the some log response parameters had the characters of obeying normal distribution laws in different measure sections. The methods and formulas, which determine the training specimens of BP never network, are given based on normal distribution. The application of this model indicates that this method is feasible and provides a good method for complex lithology.