پیش‌بینی فشار منفذی داده‌های گمانه‌های پتروفیزیکی با بهره‌گیری از روش‌های مبتنی بر هوش مصنوعی

ﻓﺸﺎر ﻣﻨﻔﺬی ﯾﮑﯽ از ﻣﻬﻢﺗﺮﯾﻦ ﭘﺎراﻣﺘﺮﻫﺎی ﻣﺨﺰﻧﯽ-ﺣﻔﺎری ﺑﻪ ﺷﻤﺎر ﻣﯽرود و آﮔﺎﻫﯽ از اﯾﻦ ﻓﺸﺎر، ﺑﺮای ﮐﺎﻫﺶ ﻫﺰﯾﻨﻪﻫﺎی ﺣﻔﺎری، اﻓﺰاﯾﺶ اﯾﻤﻨﯽ ﭼﺎه و ﭘﯿﺸﮕﯿﺮی از ﺧﻄﺮات اﺣﺘﻤﺎﻟﯽ ﺿﺮوری اﺳﺖ. اﻣﺮوزه روشﻫﺎی ﻫﻮﺷﻤﻨﺪ ﺟﺎﯾﮕﺰﯾﻦ ﺧﻮﺑﯽ ﺑﺮای ﻣﻌﺎدﻻت ﺗﺠﺮﺑﯽ ﻫﺴﺘﻨﺪ ﮐﻪ ﺗﻨﻬﺎ ﺑﺮ اﺳﺎس ﻣﺠﻤﻮﻋﻪ دادهﻫﺎی ﻣﺤﺪود، ﺟﻤﻊآوری و ﺗﻮﺳﻌﻪ ﯾﺎﻓﺘﻪ ﺑﻮدﻧﺪ. ﻣﻄﺎﻟﻌﻪ ﭘﯿﺶ رو، از 2827 ﻣﺠﻤﻮع دادهای ﻣﺮﺑﻮط ﺑﻪ ﺳﻪ ﭼﺎه از ﻣﯿﺪانﻫﺎ ﻧﻔﺘﯽ واﻗﻊ در ﺟﻨﻮب ﻏﺮب اﯾﺮان ﺑﻬﺮه ﺑﺮده اﺳﺖ. ﻣﺘﻐﯿﺮﻫﺎی ورودی ﻣﻮرداﺳﺘﻔﺎده ﺑﻪﻣﻨﻈﻮر ﭘﯿﺶﺑﯿﻨﯽ ﻓﺸﺎر ﻣﻨﻔﺬی ﺷﺎﻣﻞ 9 ﻣﺘﻐﯿﺮ ﺑﻮده ﮐﻪ ﺑﺎ اﺳﺘﻔﺎده از روش اﻧﺘﺨﺎب وﯾﮋﮔﯽ، ﺑﺮﮔﺰﯾﺪه ﺷﺪهاﻧﺪ. در اﯾﻦ ﻣﻄﺎﻟﻌﻪ از ﭼﻬﺎر اﻟﮕﻮرﯾﺘﻢ ﻫﻮش ﻣﺼﻨﻮﻋﯽ ﺷﺎﻣﻞ اﻟﮕﻮرﯾﺘﻢ ﺟﻨﮕﻞ ﺗﺼﺎدﻓﯽ1، اﻟﮕﻮرﯾﺘﻢ رﮔﺮﺳﯿﻮن ﺑﺮدار ﭘﺸﺘﯿﺒﺎن2، اﻟﮕﻮرﯾﺘﻢ ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ3 و اﻟﮕﻮرﯾﺘﻢ درﺧﺖ ﺗﺼﻤﯿﻢ4 ﺑﻪﻣﻨﻈﻮر ﭘﯿﺶﺑﯿﻨﯽ ﻓﺸﺎر ﻣﻨﻔﺬی اﺳﺘﻔﺎدهﺷﺪه اﺳﺖ. ﻧﺘﺎﯾﺞ ﻧﺸﺎن ﻣﯽدﻫﺪ ﮐﻪ دﻗﺖ ﻋﻤﻠﮑﺮد اﻟﮕﻮرﯾﺘﻢ درﺧﺖ ﺗﺼﻤﯿﻢ، ﺑﯿﺸﺘﺮ از ﺳﻪ اﻟﮕﻮرﯾﺘﻢ دﯾﮕﺮ اﺳﺖ. ﻣﻘﺪار ﻣﺮﺑﻊ ﺿﺮﯾﺐ ﻫﻤﺒﺴﺘﮕﯽ 5 (𝑅^2 ) و ﻣﺠﺬور ﻣﯿﺎﻧﮕﯿﻦ ﻣﺮﺑﻌﺎت ﺧﻄﺎ6 (rmse ) ﺑﺮای ﺗﺨﻤﯿﻦ ﻓﺸﺎر ﻣﻨﻔﺬی ﺗﻮﺳﻂ اﻟﮕﻮرﯾﺘﻢ درﺧﺖ ﺗﺼﻤﯿﻢ ﺑﻪ ﺗﺮﺗﯿﺐ 0/9985 و psi14/460ﺑﺮآورد ﺷﺪ. ازﺟﻤﻠﻪ ﻣﺰﯾﺖﻫﺎی اﯾﻦ اﻟﮕﻮرﯾﺘﻢ، اراﺋﻪ ﺑﻬﺘﺮﯾﻦ ﻧﺘﯿﺠﻪ ﺑﺪون ﻧﯿﺎز ﺑﻪ داﻧﺶ آﻣﺎری، ﺟﺪا ﮐﺮدن دادهﻫﺎی ﻏﯿﺮﺿﺮوری، آﻣﺎدهﺳﺎزی دادهﻫﺎ در زﻣﺎن ﮐﻮﺗﺎه و ﮐﺎﻫﺶ ﺧﻄﺎی ﻧﺴﺒﯽ ﺑﺎ ﯾﺎﻓﺘﻦ ﮔﺮه اﺻﻠﯽ ﺗﺼﻤﯿﻢ ﮔﯿﺮ اﺳﺖ. ﺑﻨﺎﺑﺮاﯾﻦ ﻣﯽﺗﻮان ﭼﻨﯿﻦ ﺗﺤﻠﯿﻞ ﻧﻤﻮد ﮐﻪ ﺑﺎ ﺗﻮﺳﻌﻪ اﯾﻦ روش ﻫﻮﺷﻤﻨﺪ، ﺑﺮای ﺗﻌﺪاد دادهﻫﺎی ﮐﻢ از ﻫﺮ ﻣﯿﺪان، دﻗﺖ ﻋﻤﻠﮑﺮد ﺑﺎﻻﯾﯽ ﻧﺘﯿﺠﻪ ﻣﯽﺷﻮد.

comparison of artificial intelligence algorithms to predict pore pressure using petrophysical log data

pore pressure is one of the most important reservoir-drilling parameters and knowledge of this pressure is essential for drilling costs, well safety and prevention of potential hazards. research has shown that experimental equations have good performance accuracy only for certain regions. most of these experimental equations have been compiled and developed based on a limited data set. therefore, these correlations are valid in the range of changes in the parameters of those fields and are not valid for other areas. therefore, artificial intelligent methods have given way to empirical equations. in this study, 2827 data related to three wells from one of the oil fields located in the southwest of iran have been used. the input variables used in this paper to predict the pore pressure include 9 variables that have been selected using the feature selection method. in this study, 4 artificial intelligence algorithms include; random forest algorithm, support vector regression algorithm, artificial neural network algorithm and decision tree algorithm have been used to predict the pore pressure. after reviewing the results, it was found that the performance accuracy of the decision tree algorithm is higher than the other three algorithms (performance accuracy for the entire data set including r2 = 0.9985 and rmse = 14.460 psi). among the advantages of this algorithm compared to other algorithms are the best results without the need for statistical knowledge, separation of unnecessary data, short time to prepare data and reduction of relative error by finding the main node of the decision maker and analyzing it. therefore, it can be concluded that with the development of this technique, it is possible to have high performance accuracy for a small amount of data from each field.