应用BP神经网络预测煤层含气量分布

煤层含气量分布和煤层气富集规律已成为当今重要的科研课题.充分利用BP神经网络具有的非线性映射能力、泛化能力和容错能力,批量数据处理使整体偏差值最小,分组拟合比单个样本拟合效果好的特点预测煤层气含量;煤层含气量在一定范围内变化,于是把煤层含气量作为因变量的数据进行分级预测,并且适当调节样本允许误差且允许个别错误存在,以减少模型整体误差.煤层的埋深、镜质组、灰分和挥发分为影响QP区块的主要因素,将这4个影响因素作为变量建立BP神经网络模型,调节网络模型各项参数,分配不同学习训练样本、检验样本和坚持样本以找出合理的神经网络学习训练结构.再与地质统计学和克里金插值法有机结合来预测煤层含气量分布规律和探索煤层气富集规律.

Prediction of the Distribution of the Coal Reservoir Gas Content Based on BP Neural Network

It is current important question to predict distribution of coal reservoir gas content and the enrichment law of coal bed methane. When BP neural network which has some abilities including nonlinear, generalization and fault - tolerant is used for simulation, it is that the data are divided into groups better than the single, so coal reser- voir gas content data are divided into four groups as dependent variable. There are many factors effect on the gas content in QP Region Qinshui Basin, especially, depth, vitrinite, volatile and ash are main factors in the region, which are used as independent variables in the BP neural network. The reasonable network is found by the setting relational parameters and the data are divided into the training, test and persistence samples. Then the distribution of the coal reservoir gas content is predicted by the right model and the map of coal reservoir gas content is drawn by using geostatistics and interpolation. Then the distribution of coal reservoir gas content and the enrichment law of coal bed methane have been showed.