神经网络及模式识别技术在测井解释中的应用

Applications of Numerical Calculation and Model Identification Technology Based on Neural Network in Log Interpretation.

针对四川碳酸盐岩储层实际情况 ,用测井资料与岩心分析结果组成孔隙度、含水饱和度样本 ,经BP网络训练后得到模型参数。使用模型参数进行孔隙度、含水饱和度计算 ,计算结果精度高。BP网络在产能评价方面也有明显优势 ,能较好地表达储层参数与产能之间的关系 ,提高预测结果精度。对汉明网络结构作了适当的调整 ,使其适用于输入为连续值的模式识别问题。在储层流体性质判别方面 ,气层、水层的判别符合率达 93% ;该网络在多个地区的测井相分析中应用表明 ,能提高沉积微相识别率和预测符合率。实际应用证实 ,神经网络技术能提高测井解释中的数值计算精度和模式识别符合率 。

According to Sichuan carbonate reservoir, porosity and water saturation samples are constituted by using logging data and core analytic results, and the model parameters for calculating the pore and water saturation are obtained after BP network training. Its accuracy is high. The BP network is obviously superior in productivity evaluation. It can not only express the relation between reserovir parameters and reservoir output, but also enhance precision of forecast results. Hamming network was adjusted to suit the pattern recognition with successive input values. In identification of reservoir liquid property, its agreement ratio is 93 percent. Application of such a network in analysis of many zone′s logging facies shows it can increase identifying ratio, and improve calculation accuracy of numerical value and model identifying agreement. Now it is applied to fine log interpretation and reserve production calculation.