改进 GＲNN 网络预测致密砂岩气层压裂产能

致密砂岩储层孔隙度小、渗透率低、含气饱和度低，基本上没有自然产能，需要进行压裂，所以压裂产能的预测很重要。广义回归神经网络 ( GＲNN) 稳定，对样本数量的要求低。产能预测关键是样本的选取以及扩展因子的选取。在原有的 GＲNN 预测产能的基础上，利用交叉验证法改进 GＲNN 网络，选取最优的样本确定最优的 GＲNN 网络结构，利用循环判断法，选取最优的扩展因子。改进的 GＲNN 神经网络可以避免确定 GＲNN 网络结构和扩展因子过程中过多的人为影响。笔者利用灰色关联分析法分析压裂产能的影响因素，利用改进的 GＲNN 网络有针对性地建立适合苏里格地区致密砂岩气层的压裂产能预测模型。结果表明该方法在苏里格地区气层压裂产能预测中有较好的应用效果。

Prediction of gas productivity based on improved GＲNN for post-frac tight sandstone reservoirs

Tight sandstone reservoirs are always characterized by low porosity，low permeability and low gas saturation，almost with no natural capacity，which is requested fracturing for productivity． Therefore the fracturing capacity prediction is very necessary． GＲNN neural network is stable，with low demand for the number of samples． The key is the selection of samples and the expansive factor in production prediction of GＲNN． The authors apply the cross-validation method to select the samples to determine the optimal GＲNN network structure，and use the re- cycled judgment to select the optimal expansive factor on the basis of intrinsic GＲNN productivity prediction． The improved GＲNN neural network could avoid the human impact in the selection of net structure and spreading factor． Taking the gray correlation analysis method，the authors determine the fracturing capacity factors，then use the im- proved GＲNN network to predict the gas production for post-frac tight sandstone reservoirs in the Sulige area． The results show that the method is well in application in the gas production prediction for tight sandstone reservoirs in the Sulige area．