fa ارائه یک الگوریتمی مبتنی بر یادگیری عمیق برای تخمین شاخص شکنندگی با استفاده از داده‌های لاگ‌های متداول در سازند آسماری یکی از میادین نفتی جنوب‌غرب ایران ژئوفیزیک مهندسی Engineering Geophysics مقاله پژوهشی Original Research <div style="text-align: justify;"><span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;"><span style="line-height:2;"><span dir="RTL" lang="FA">شاخص شکنندگی یکی از پارامترهای مهم در بررسی&shy; و مدل&shy;سازی‌&shy;های&shy; ژئومکانیکی است. روش‌&shy;های زیادی برای تخمین شاخص شکنندگی ارائه شده است. یکی از روش&shy;‌هایی که امروزه زیاد مورد استفاده قرار می‌&shy;گیرد روش‌&shy;های هوشمند است. در این مطالعه هدف ارائه الگوریتمی جدید با استفاده از الگوریتم‌&shy;های یادگیری عمیق جهت پیش&shy;بینی شاخص شکنندگی در یکی از چاه&shy;‌های میدان هیدروکربنی در جنوب غرب ایران می‌&shy;باشد. در این مقاله ابتدا پارامترهای موثر برای ورودی الگوریتم‌&shy;ها با استفاده از ضریب همبستگی پیرسون مشخص گردید و در ادامه با استفاده از (شبکه عصبی بازگشتی + شبکه عصبی پرسپترون چندلایه) </span>(</span>LSTM+MLP<span style="line-height:2;">)&nbsp;<span dir="RTL" lang="FA">و (شبکه</span>&shy;<span dir="RTL" lang="FA"> عصبی تبدیلی + شبکه عصبی بازگشتی) </span>(</span>CNN+ LSTM<span style="line-height:2;">)&nbsp;<span dir="RTL" lang="FA">شاخص شکنندگی تخمین زده شد و مقدار خطا </span>(</span>MSE<span style="line-height:2;">)&nbsp;<span dir="RTL" lang="FA">و ضریب تعیین </span>&nbsp;(</span>R²<span style="line-height:2;">)&nbsp;<span dir="RTL" lang="FA">برای داده‌&shy;های آموزش و تست محاسبه گردید که برای داده‌&shy;های آموزش و تست هر دو الگوریتم دارای ضریب تعیین نزدیک به 1 و خطای بسیار کم به دست آمده است. همچنین جهت اطمینان از نتایج الگوریتم</span>&shy;<span dir="RTL" lang="FA">ها بخشی از داده به عنوان داده کور کنار گذاشته شد و خطا و ضریب تعیین برای این</span> <span dir="RTL" lang="FA">داده&shy;‌ها نیز محاسبه</span> <span dir="RTL" lang="FA">گردید که خطا (</span></span></span></span></span><span style="font-family:yekanYW;"><span style="color:#000000;"><span style="font-size:14px;"><m:r>MSE</m:r> <m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr>_CNN+LSTM</m:r> <m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr>=26.0425,&nbsp; </m:r> <m:ctrlpr></m:ctrlpr> <m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr>MSE</m:r> <m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr>_LSTM+MLP</m:r> <m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr>=</m:r><m:r>32.07</m:r></span><span style="font-size:10.0pt"><m:r>5</m:r></span></span></span><span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;"><span style="line-height:2;"><span dir="RTL" lang="FA">)&nbsp;به دست آمده است&nbsp; و&nbsp; ضریب تعیین<br> (</span></span></span></span></span><span style="font-size:10.0pt"><span cambria="" math="" style="font-family:"><m:ctrlpr></m:ctrlpr></span></span><span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span></span><span style="font-size:10.0pt"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr><span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;">R²</span></span></span></m:r></span></span><span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span></span>_{<span style="font-size:8px;"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr><span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;">_CNN+LSTM</span></span></span></m:r></span></span>}<span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;"> &nbsp;</span></span></span><span style="font-size:10.0pt"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr><span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;">=0.8064,&nbsp; </span></span></span></m:r></span></span><span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span></span><span style="font-size:10.0pt"><span cambria="" math="" style="font-family:"><m:ctrlpr></m:ctrlpr></span></span><span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span></span><span style="font-size:10.0pt"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr><span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;">R²</span></span></span></m:r></span></span><span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span></span>_{<span style="font-size:8px;"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr><span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;">_LSTM+MLP</span></span></span></m:r></span></span>}<span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;"> &nbsp;</span></span></span><span style="font-size:10.0pt"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr><span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;">=0.7615</span></span></span></m:r></span></span><span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;"><span style="line-height:2;"><span dir="RTL" lang="FA">) به دست آمده است. نتایج بیانگر کارآیی الگوریتم‌&shy;های یادگیری عمیق معرفی شده به عنوان روشی<br> جدید در پیش&shy;بینی شاخص شکنندگی می‌&shy;باشد که در مقایسه دو الگوریتم ارائه شده،الگوریتم&nbsp; (</span></span></span></span></span><span style="font-size:10.0pt"><span new="" roman="" style="font-family:" times=""><span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;">C</span></span><span style="font-family:yekanYW;"><span style="font-size:14px;">NN+LSTM</span></span></span></span></span><span style="color:#000000;"><span style="font-size:14px;"><span style="font-family:yekanYW;"><span style="line-height:2;"><span dir="RTL" lang="FA">) دارای دقت بالاتر و خطای کمتری می‌&shy;باشد.</span></span></span></span></span><br> <br> <i><span style="font-size:11.0pt"><span cambria="" math="" style="font-family:"><m:r>&nbsp;</m:r></span></span></i> <span style="font-size:12.0pt"><span new="" roman="" style="font-family:" times=""> </span></span></div> <div style="text-align: justify;"><span style="line-height:2;"><span style="font-size:14px;"><span style="font-family:yekanYW;"><span style="color:#000000;">The brittleness</span><span style="color:#000000;"><span lang="EN">index is one of the most important parameters in geomechanical analysis and modeling. Many methods have been proposed to estimate the </span>brittleness <span lang="EN">index. One of the recently used methods is the &nbsp;intelligent method. In this paper, firstly the aim is to introduce a new algorithm using deep learning algorithms to predict the </span>brittleness <span lang="EN">index in one of the wells of the hydrocarbon field in southwest Iran. In this article, first, the effective features for the input of the algorithms were determined using Pearson&#39;s correlation coefficient, and then using </span>(<span lang="EN">recurrent neural network + multi-layer perceptron neural network) (LSTM + MLP) and (convolutional neural network + recurrent neural network) (CNN+ LSTM) </span>brittleness <span lang="EN">index was estimated and the mean error value (MSE) and coefficient of determination (R²) were calculated for the training and test data. For both training and test data, both algorithms have a coefficient of determination close to 1 and a very low error. Also, in order to ensure the results of the algorithms, a part of the data was set aside as blind data, and the error and coefficient of determination were calculated for this data, and the error was </span> </span></span></span></span><span style="color:#000000;"><span style="line-height:2;"><span style="font-size:9.0pt"><span cambria="" math="" style="font-family:"><m:ctrlpr></m:ctrlpr></span></span><span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span><span style="font-size:9.0pt"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr><span style="font-size:14px;"><span style="font-family:yekanYW;">MSE</span></span></m:r></span></span><span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span>_{<span style="font-size:9.0pt"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr>_{<span style="font-size:14px;"><span style="font-family:yekanYW;">CNN+LSTM</span></span>}</m:r></span></span>}<span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span><span style="font-size:9.0pt"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr><span style="font-size:14px;"><span style="font-family:yekanYW;">=26.0425,&nbsp; </span></span></m:r></span></span><span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span><span style="font-size:9.0pt"><span cambria="" math="" style="font-family:"><m:ctrlpr></m:ctrlpr></span></span><span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span><span style="font-size:9.0pt"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr><span style="font-size:14px;"><span style="font-family:yekanYW;">MSE</span></span></m:r></span></span><span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span>_{<span style="font-size:9.0pt"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr>_{<span style="font-size:14px;"><span style="font-family:yekanYW;">LSTM+MLP</span></span>}</m:r></span></span>}<span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span><span style="font-size:9.0pt"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr><span style="font-size:14px;"><span style="font-family:yekanYW;">=32.0751</span></span></m:r></span></span><span style="font-size:14px;"><span style="font-family:yekanYW;"> <span style="position:relative"><span style="top:2.5pt"> </span></span><span lang="EN">&nbsp;and the coefficient of determination was </span> </span></span><span style="font-size:9.0pt"><span cambria="" math="" style="font-family:"><m:ctrlpr></m:ctrlpr></span></span><span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span><span style="font-size:9.0pt"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr><span style="font-size:14px;"><span style="font-family:yekanYW;">R²</span></span></m:r></span></span><span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span>_{<span style="font-size:9.0pt"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr>_{<span style="font-size:14px;"><span style="font-family:yekanYW;">CNN+LSTM</span></span>}</m:r></span></span>}<span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span><span style="font-size:14px;"><span style="font-family:yekanYW;">&nbsp;</span></span><span style="font-size:9.0pt"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr><span style="font-size:14px;"><span style="font-family:yekanYW;">=0.8064,&nbsp; </span></span></m:r></span></span><span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span><span style="font-size:9.0pt"><span cambria="" math="" style="font-family:"><m:ctrlpr></m:ctrlpr></span></span><span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span><span style="font-size:9.0pt"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr><span style="font-size:14px;"><span style="font-family:yekanYW;">R²</span></span></m:r></span></span><span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span>_{<span style="font-size:9.0pt"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr>_{<span style="font-size:14px;"><span style="font-family:yekanYW;">LSTM+MLP</span></span>}</m:r></span></span>}<span style="font-size:14px;"><span style="font-family:yekanYW;"> </span></span><span style="font-size:14px;"><span style="font-family:yekanYW;">&nbsp;</span></span><span style="font-size:9.0pt"><span cambria="" math="" style="font-family:"><m:r><m:rpr><m:scr m:val="roman"><m:sty m:val="p"></m:sty></m:scr></m:rpr><span style="font-size:14px;"><span style="font-family:yekanYW;">=0.7615</span></span></m:r></span></span><span style="font-size:14px;"><span style="font-family:yekanYW;"> <span style="position:relative"><span style="top:2.5pt"> </span></span><span lang="EN">.</span><span lang="EN"> The results show the effectiveness of the introduced deep learning algorithms as a new method in predicting the </span>brittleness <span lang="EN">index, and comparing the two algorithms presented, the CNN+LSTM algorithm has higher accuracy and less error.</span></span></span></span></span></div> شاخص شکنندگی, داده‌های پتروفیزیکی, یادگیری عمیق, ارزیابی مدل brittleness index, petrophysical data, deep learning, evaluation model 341 364 http://jeg.khu.ac.ir/browse.php?a_code=A-10-1992-1&slc_lang=fa&sid=1 Farhad Mollaei فرهاد ملائی mollaeifarhad@ut.ac.ir 10031947532846006665 10031947532846006665 No Tehran Univercity دانشگاه تهران Reza Mohebian رضا محبیان mohebian@ut.ac.ir 10031947532846006666 10031947532846006666 Yes Tehran Univercity دانشگاه تهران Ali Moradzadeh علی مرادزاده a_moradzadeh@ut.ac.ir 10031947532846006667 10031947532846006667 No Tehran Univercity دانشگاه تهران