计算机应用
計算機應用
계산궤응용
COMPUTER APPLICATION
2014年
z2期
186-189
,共4页
油页岩%含油率%测井参数%BP神经网络%测井解释模型
油頁巖%含油率%測井參數%BP神經網絡%測井解釋模型
유혈암%함유솔%측정삼수%BP신경망락%측정해석모형
oil shale%oil content%log parameter%BP neural network%log interpretation model
根据测井资料计算油页岩含油率多采用△logR法或改进的△logR法，这些方法中参数获取过程中易产生诸多误差，且这些方法是建立在油页岩含油率与特征测井曲线值是线性关系的基础上的，而在实际非均质性地层中，测井对油页岩含油率参数的响应在本质上必然是非线性的。基于此，运用BP神经网络来预测柴达木盆地北部地区侏罗纪油页岩含油率。首先分析研究区段测井数据的数理统计分布特征，在优选学习样本的基础上再采用一种基于LM（ Levenberg-Marquardt）算法的BP神经网络进行含油率预测，最后得出一组由40个连接权值与11个阈值组成的含油率参数解释模型，油页岩含油率预测值与岩心实验室分析值吻合很好，均方误差能控制在0.1918。因此，运用此模型可以预测相同地质背景条件下的油页岩含油率。
根據測井資料計算油頁巖含油率多採用△logR法或改進的△logR法，這些方法中參數穫取過程中易產生諸多誤差，且這些方法是建立在油頁巖含油率與特徵測井麯線值是線性關繫的基礎上的，而在實際非均質性地層中，測井對油頁巖含油率參數的響應在本質上必然是非線性的。基于此，運用BP神經網絡來預測柴達木盆地北部地區侏囉紀油頁巖含油率。首先分析研究區段測井數據的數理統計分佈特徵，在優選學習樣本的基礎上再採用一種基于LM（ Levenberg-Marquardt）算法的BP神經網絡進行含油率預測，最後得齣一組由40箇連接權值與11箇閾值組成的含油率參數解釋模型，油頁巖含油率預測值與巖心實驗室分析值吻閤很好，均方誤差能控製在0.1918。因此，運用此模型可以預測相同地質揹景條件下的油頁巖含油率。
근거측정자료계산유혈암함유솔다채용△logR법혹개진적△logR법，저사방법중삼수획취과정중역산생제다오차，차저사방법시건립재유혈암함유솔여특정측정곡선치시선성관계적기출상적，이재실제비균질성지층중，측정대유혈암함유솔삼수적향응재본질상필연시비선성적。기우차，운용BP신경망락래예측시체목분지북부지구주라기유혈암함유솔。수선분석연구구단측정수거적수리통계분포특정，재우선학습양본적기출상재채용일충기우LM（ Levenberg-Marquardt）산법적BP신경망락진행함유솔예측，최후득출일조유40개련접권치여11개역치조성적함유솔삼수해석모형，유혈암함유솔예측치여암심실험실분석치문합흔호，균방오차능공제재0.1918。인차，운용차모형가이예측상동지질배경조건하적유혈암함유솔。
Method of△logR and advanced method of△logR arre usually adopted to calculate oil content of oil shale with log data. These methods easily cause some errors in the process of calculating parameters, and these methods are based on linear relation between oil content and characteristic log values. However, it was absolutely a nonlinear relation between them in the actual heterogeneous stratum. Therefore, BP neural network based on LM ( Levenberg-Marquardt ) algorithm was adopted to calculate the oil content in Jurassic strata of northern Qaidam basin. Firstly, mathematical statistics distribution feature of log data were analyzed with Matlab; Ssecondly, oil content values were predicted with BP neural network based on LM algorithm after the excellent samples had been chosen; finally, a matrix composed of 40 link weights and 11 thresholds was the parameter interpretation model of oil content. Results of the BP neural network prove that theoretical calculating values match well with the core experimental measuring values, and the mean square error can be controlled within 0. 191 8. Therefore, this parameter interpretation model can be promoted in the area of the same geology background.
