基于离散过程神经网络页岩油气储层有机碳含量预测

发布时间：2018-09-13 12:22

【摘要】：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。
[Abstract]:Under the influence of stratigraphic lithology, the accuracy of (TOC) fitting for organic carbon content is low. In order to improve the accuracy of TOC fitting and to reduce the cumulative error of continuous signals in ordinary neural networks, a TOC fitting prediction model for limit learning discrete process neural networks is proposed. In the model, the process input is simulated by vector, and the time domain aggregation of discrete samples is completed by the numerical integration of parabolic interpolation. By analyzing the structure of neurons in discrete process, a limit learning training algorithm is proposed. After the parameters of hidden layer are randomly assigned, the output weights are solved by Moore-Penrose generalized inverse, and the model learning speed is fast. Finally, the method is applied to the prediction of TOC fitting. The logging curve which is most sensitive to the TOC response is selected as the feature input of the model by using the correlation analysis. Compared with the traditional method and other neural networks, this method has higher fitting accuracy and better correlation between the predicted TOC and the measured values.
【作者单位】：东北石油大学计算机与信息技术学院;中国石油大学非常规油气与新能源研究院;山东科技大学信息科学与工程学院;
【基金】：国家自然科学基金项目(41602141,41402109,41330313)
【分类号】：TP183;P618.13

【相似文献】