自构形神经网络模型在天然气储量评价中的应用初探

在ＢＰ（Ｅｒｒｏｒ Ｂａｃｋ－Ｐｒｏｐａｇａｔｉｏｎ）算法基础上，采用ＢＩ（ＢａｃｋＩｍｐｅｄａｎｃｅ）算法，运用了自组织优化隐层节点数和自动化优化网络因子的方法，使得人工神经网络ＡＮＮ的计算速度、精度和柔韧性有所提高，且在微机上其操作变得更加容易。在勘探成熟的气藏中，按照天然气成藏理论，选取能够系统反映气藏的８个储量评估参数，进行网络学习，建立储量评估模型。应用所建的网络模型对正处于勘探阶段的气

APPLICATION OF IMPROVED SELF-CONFIGURING ARTIFICIAL NEURAL NETWORKS lO RESERVES ESTIMATION

Based on the BP(Error Back Propagation) algorithm,the authors applied the improved BI(Back Impedance Neural Networks) algorithm and put forward the method of self configuring nodes of hidden layer and automatic optimization for ANN factors.According to the relationship factors among hidden nodes at every running time, the ANN automatically decided which hidden nodes would be deleted or incorporated. Judging from the variation of absolute errors between the previous running time and this running time,the ANN could also automatically adjust ANN factors to make it quickly stable and convergent.By above methods the calculating speed and calculating accuracy of ANN is raised, and it's easier to operate on the personal computer(PC) than the normal algorithm of ANN.The authors selected eight parameters, about natural gas reserves of reservoirs explored to maturity,as the trained samples to build the ANN model for estimating the ranks of reserves.The ANN model was made up of eight input layer nodes,ten hidden layer nodes and three ouput layer nodes. Then by which the latest exploring reservoirs were well classified according to their reserves. Since it had a high calculation accuracy,the authors obtained certain interesting result when calculating the reserves by it. By comparison the above results obtained by ANN and the actual proved reserves by experts were similar. So this paper presents a new method for reserves estimation in the nature gas exploration.