改进BP神经网络模型在长庆气田产能预测中的应用

Application of Neural Network Technique for Productivity Evaluation in Changqing Gasfield

在改进的神经网络训练算法的基础上,提出了利用神经网络快速预测储层产能的方法步骤。长庆气田属于非常规碳酸盐岩气藏,地质条件复杂,结构特征多样,产能影响因素较多,且各种因素交互重叠,用常规方法进行产能预测,其精度远远不够。为了迅速、准确地判断储层的性质,选用了BP神经网络,利用测井参数,建立了长庆气田产能预测模型,提高了预测的精度。常规神经网络在训练样本时分母项易趋于0,导致运算进入死循环,降低了结果的可信度。改进的神经网络模型克服了上述现象,具有绝对收敛性,且隐含层的神经元个数容易调整。将该改进模型用于储层产能预测,正确率达94%以上。

On the basis of improving training algorithm of neural networks,procedures for quickly predicting the reservoir productivity were proposed by using neural network.Changqing Oilfield belonged to non-conventional carbonate gas reservoirs where geologic conditions were complex with various structural characters and productivity influenced factors,it was unsufficient to predict the productivity with conventional methods.For quickly and accurately predicting the reservoir property,neural networks were selected and well logging parameters were used to establish a productivity predicting model for improving the predicting accuracy in Changqing Oilfield.In conventional method,the nominator was approached to zero,dead cycle was caused and credibility of result was reduced.The shortcomings are avoided by using the improved neural network,absolute convergence is ensured,the cell number of hidden layer is easy to be adjusted.The model is used for reservoir productivity prediction with correcting rate over 94%.