基于改进BP神经网络的页岩吸附量预测模型

摘要

Shale adsorption is a complex process involving multiple factors such as organic carbon content(TOC),organic matter m aturity(Ro),temperature(T) and pressure(P).Shale adsorption cannot be effectively analyzed by the conventional adsorption isotherm model in the case of multi-factor simultaneous effect and random coupling.Therefore,four major influential factors of shale adsorption were analyzed and summarized in this study.The prediction model of the shale adsorption capacity was built based on the BP neural network that was improved by genetic algorithm and the kernel density estimation algorithm,whose optimal number of the hidden layer nodes was determined by trial and error method.BP network was trained and tested by the 80 sets of experimental data from 4 samples under different conditions.The results show that the relative error of 75 sets of train data is in the range of-2％-2％,the average relative error is 2.48％;the relative error of 5 sets of test data is in the range of-10％-10％,the average relative error is 4.999％;the correlation coefficient between the predicted value and the actual value is 0.978 9.In addition,the sensitivity of 4 influencing factors to the adsorption capacity was analyzed by orthogonal test,and the sensitivity order is p ＞T＞TOC＞R..It is concluded that the model has high prediction accuracy to analyze shale adsorption under multi-factor simultaneous effect.%页岩吸附是受TOC(总有机碳质量分数)、Ro(镜质体反射率)、T(温度)和p(压力)等多重因素共同作用的复杂过程,常规的等温吸附模型无法对多因素同时作用及随机耦合的页岩吸附情况进行分析.为此,文中总结分析了影响页岩吸附能力的主要影响因素,通过遗传算法和核密度估计法改进了BP神经网络,利用试错法确定隐含层最优节点数为12,从而构建了吸附量预测模型.基于4个岩样、80组吸附实验数据,对BP网络进行训练和检验.其中:75组训练数据的预测相对误差处于±2％,平均相对误差为2.480％;5个检验组的预测相对误差处于±10％,平均相对误差为4.999％,预测值与实际值的相关系数为0.978 9.另外,利用正交试验分析了4个影响因素对吸附量的敏感性,发现其敏感性排序为p＞T＞ TOC＞Ro.实验结果表明,该模型具有较高预测精度,可用于分析多因素同时作用下的页岩吸附情况.