Jeofizik yer fiziği olarak adlandırılıp, üzerinde yaşadığımız dünyanın, fiziğin temel ilkelerinden yararlanarak araştırıp bu ilkeler doğrultusunda amaca uygun olarak doğal yeraltı kaynaklardan yarar sağlamak veya yer kaynaklı doğal afetlerle ilişkili olabilecek yapıların ortaya çıkarılmasını hedeflemektedir. Bu bağlamda yeraltı araştırmalarında yeryüzünde araştırmaya konu olmuş doğal kaynak veya tektonik yapıların modellenip ortaya çıkarmasını, fiziksel problemlere dayandırarak ortaya bir model koyan bir bilim dalıdır. Jeofizik araştırma yöntemleri, yer üstünde ve yer altında olmak üzere iki türlü inceleme alını bulunmaktadır. Çalışmada sondaj kuyusundan alınan Jeofizik kuyu logu ölçü değerleriile sondajdan alınan numunelerin laboratuvar değerlerinin sonuçları arasındaki ilişkiler,geleneksel ve esnek hesap yöntemlerinden faydalanarak yapılmaya çalışılacaktır. Çalışmada uranyum sahasında açılan 290 kuyudan 130 sondaj kuyusundan elde edilen466 adet veri seti kullanılmıştır. Bu veri seti sondajdan alınan jeofizik ölçüde elde edilen Gamma Ray ve Uranyum sayısal değerleri ile laboratuvar ortamında elde edilen Uranyum sayısal değerlerinden oluşmaktadır. Laboratuvar değerleri çıktı olarak kabul edilip jeofizik ölçüde elde edilen Gamma Ray ve Uranyum sayısal değerleri girdi olarak alınmıştır. Girdilerin birbiriyle kombinasyonu sonucunda 3 farklı grup oluşturulmuştur. Oluşan grupların veri seti kullanılarak geleneksel yöntem olan doğrusal regresyon analizi için 3 model oluşturulmuştur. Esnek hesap yöntemlerinden yapay Sinir Ağları (YSA) yöntemi için 30 model ve Adaptif Sinirsel Bulanık Çıkarım Sistemi (ANFİS) yöntemi için ise 36 model olmak üzere toplamda 69 modelden oluşan bir veri seti eğitilip test işlemlerine tabi tutulmuşlardır. Modeller incelendiğinde ANFİS'de kullandığımız modellerin tahmin performans sonuçlarının diğer yöntemlere göre daha iyi sonuçlar verdiği gözlenmiştir.
Geophysics is called earth physics, and it aims to investigate the world we live on by using the basic principles of physics and to benefit from natural underground resources in accordance with these principles or to reveal structures that may be associated with ground-based natural disasters. In this context, it is a branch of science that models and reveals the natural resources or tectonic structures that have been the subject of research on earth in underground researches, based on physical problems. There are two types of study areas: geophysical research methods, above ground and underground. Geoscientists narrow down the field based on above-ground signs of any underground-related search. Then, the narrowed ground is examined with geophysical methods and the physical values of the material sought are found and its location is tried to be determined. Drilling, which is the last step for the accuracy of the ground, which is not seen above the ground but determined by physical values with geophysical methods, is done. Since the drilling to be done is very laborious and expensive, it is tried to get all the information about the underground that has been drilled to the maximum extent.The acquisition of this information starts with the geophysical well log measure, which gives us the physical properties of the cut formations as soon as the well is finished. Continues with the evaluations of the cores taken while drilling. In this study, the relationships between the underground investigation in geophysics, that is, the geophysical well log measurement values taken from the borehole and the laboratory values of the core sample obtained by drilling, will be examined. This relationship is extremely important for us researchers as it is directly related to the formations cut in the well and the ores cut in the well and their thickness. The perusal will be made by making use of the traditional and flexible calculation methods that are frequently used today. Geophysical gamma ray (JGR), which is our geophysical measurement values obtained in drilling and laboratory uranium (LU) values obtained from core samples taken from ore levels cut while drilling with geophysical uranium (JU) will be processed. However, due to the negativities that may occur during drilling (in case of negativity that will occur in the core purchase or when there is no core purchase). The number of wells drilled in the study area by the General Directorate of Mineral Research and Exploration, and the number of wells we have obtained permission to use in our thesis, is 290. Among the wells whose data we obtained permission to use, 130 wells with the data we used in our study were determined. While making the drilling selection, features such as the parameters (geophysical measurements and laboratory results) we used in the drilling data and the fact that ore was cut during the drilling were taken into account. 466 data sets were obtained from 130 boreholes from 466 levels. This data set is a triple data set consisting of the geophysical gamma ray and geophysical uranium numerical values obtained from the geophysical measure obtained from the drilling and the Uranium numerical values obtained in the laboratory environment. Since today's researchers accept the laboratory results of the core samples obtained by drilling as the final result, laboratory uranium values are accepted as output, that is, the geophysical gamma ray and geophysical uranium numerical values obtained in the geophysical scale are taken as inputs. Out of 466 datasets obtained from geophysical well measurements and laboratory results, 335 datasets were randomly divided into two as 131 datasets of test data to control the results of training and training data. As a result of the combination of the obtained data with each other, 3 different groups were formed. By using the training data of the groups formed, the traditional method of linear regression analysis and flexible calculation methods Artificial Neural Networks (ANF) and Adaptive Neural Fuzzy Inference System (ANFIS) training processes were carried out. In order to evaluate the prediction performance of the models created in the training process, traditional (regression analysis) and flexible calculation methods (ANF and ANFIS) were applied to the same models using test data. In the traditional method, 3 models were created using simple linear regression analysis. In flexible calculation methods, feed forward back propagation (IBGY) method in YSA and Levenberg-Marquardt (LM) learning algorithm in network training were used. In the hidden layer, a total of 30 ANN models were created, with the number of neurons (N) increasing from 2 to 20 in pairs. Triangular (tri), trapezoidal (trap), generalized bell shaped (gbell) and gaussian (gaussian) membership functions and membership function cluster number in ANFIS to examine the effect of membership function type (Membership Function, MF) and membership function cluster number on model performance. As a result, a total of 36 ANFIS models were created as 2,3,4, respectively. In the models created by using traditional and flexible calculation methods, laboratory values were estimated by using geophysical well log measure taken from drilling. In order to evaluate the efficiency of the model, coefficient of determination (R²) and Mean Squared Error (MSE) criteria were used. In the models created by using traditional and flexible calculation methods, geophysical well log taken from drilling is trained using numerical values, the estimation performances of the laboratory values were examined by passing the test phase. In order to evaluate the prediction performance efficiency of the models, an evaluation was made using the coefficient of determination (R²) and the mean squared error (Mean Squared Error, MSE) criteria. The first criterion in forecast performance was the coefficient of certainty. If the coefficient of determination is equal, the estimation performance is tried to be made by considering the mean square error criterion. The results of 69 models created for the study were examined one by one. Considering the results of the criteria used for estimation performance, the estimation performance of the regression analysis used in traditional methods is relatively low. It was observed that ANN and ANFIS had higher estimation performances than regression analysis. Among the flexible calculation methods, it has been observed that the ANFIS estimation performances are slightly better than the ANN method. As a result, the estimation performances of the ANN and ANFIS methods, which will be applied to the measured values of the radioactive geophysical well log, have shown us that the results are close to the laboratory values.
Except where otherwise noted, this item's license is described as http://creativecommons.org/licenses/by/4.0/
