基于离散过程神经网络页岩油气储层有机碳含量预测

受地层岩性变化影响,传统方法进行有机碳含量(TOC)拟合预测精度偏低。为提高TOC拟合精度,减小普通神经网络对连续信号的时间累积误差,提出一种极限学习离散过程神经网络的TOC拟合预测模型。模型用向量模拟过程式输入,内部通过抛物插值的数值积分完成离散样本的时域聚合。通过对离散过程神经元的结构分析,提出极限学习训练算法,在隐层相关参数随机赋值后,通过Moore-Penrose广义逆求解输出权值,模型学习速度快。最后将该方法应用于TOC拟合预测,利用相关性分析,选取对TOC响应最敏感的测井曲线作为模型的特征输入。与传统方法和其他神经网络对比,该方法的拟合精度较高,预测TOC与实测值有更好的相关性。

Total organic carbon content prediction of shale reservoirs based on discrete process neural network

Traditional methods in TOC fitting generally have low precision due to the effects of lithology change. In order to improve TOC fitting precision and to reduce the time cumulative error for continuous signals in the artificial neuron network, an extreme learning discrete process neural network is proposed. A vector form is used to simulate the process input in the model. The time domain aggregation for discrete data input is controlled by the parabolic interpolation using numerical integration in the discrete process neuron. Through analysis of structure of discrete process neuron, an extreme learning algorithm is proposed. The parameters of the hidden layer are randomly assigned and the Moore-Penrose generalized inverse is used to compute the output weights. The method is applied to TOC fitting and prediction usingsome logging curves which have most sensitive response for TOC. The TOC fitting results are compared with the traditional methods and other neural network. The results show that the proposed method has higher fitting precision and faster learning speed, and the predicted TOC and actual TOC have better correlations.