BP神经网络在原煤固定碳含量预测中的应用

Application of BP Neural Network to the Content Prediction of Fixed Carbon in Raw Coal

固定碳是煤层工业组分的重要参数之一,传统的均质理论、线性假设在储层参数预测过程中存在着模型简单、受地层非均质性影响较高等方面的不足,参数预测结果误差较大。为提高对工业组分预测的精度及推广能力,采用非线性关系的BP神经网络方法,通过测井数据预处理、挑选学习样本等工作建立预测原煤固定碳含量的BP神经网络模型。经过学习效果检验及误差分析得出,由BP神经网络预测得到的原煤固定碳含量与实验分析数据之间误差小,预测精度高,具有较好的可推广性。

Fixed carbon is one of the important parameters of coal industrial components,some deficiencies exist during the prediction of reservoir parameters with the traditional homogeneity theory and linear hypothesis because of the simple model and the high impact of formation heterogeneity,leading to a big error in parameter prediction. To improve the prediction precision and generalization ability for industrial components,the BP neural network method with nonlinear relationships was taken. A BP neural network model was built to predict the fixed carbon content of raw coal through the work of data preprocessing,learning samples selection and so on. After the learning effect test and error analysis,it shows that the error between predicted content of raw coal fixed carbon by BP neural network and that by experimental measurement is low,the prediction accuracy is high and this method has a good generalization.