Heping Pan

Prediction of porosity in crystalline rocks using artificial neural networks: An example from the Chinese Continental Scientific Drilling Main hole

Porosity plays an important part of understanding permeability and fluid flow within the continental, crystalline rocks. Geophysical well logs are presently the most consistent means of providing continuous information for porosity estimation. However, it is difficult to interpret geophysical well logs data in crystalline rocks due to their complex geological features and the difficulty in understanding and using the complex and intensive information content in these data. Motived by the successful prediction abilities of artificial neural networks (ANN) to solve different problems in geophysics, this study explore the applicability of using ANNs to predict porosity in continental, crystalline rocks. This ANN technique is calibrated on Chinese Continental Scientific Drilling Main Hole (CCSD-MH) data, which provides core porosity data combined with four geophysical well logs (density, neutron porosity, sonic and resistivity). The data from CCSD-MH is utilized to train feed-forward backpropagation (FFBP) neural network and radial basis function (RBF) neural network to derive a relationship between geophysical well logs and porosity, and hence predict porosity accurately. The findings demonstrate that ANNs provide better performances with sets of three geophysical well logs (density, sonic and resistivity) than regression technique. Comparison of FFBP to RBF showed that RBF reveals better stability and more accurate performances than FFBP. Based on the success achieved in this study, this intelligence artificial technique can be a very advantageous tool in facilitating the task of geophysicists in the framework of research drillings in continental crust.

Lithology and mineralogy recognition from geochemical logging tool data using multivariate statistical analysis

The availability of a deep well that penetrates deep into the Ultra High Pressure (UHP) metamorphic rocks is unusual and consequently offers a unique chance to study the metamorphic rocks. One such borehole is located in the southern part of Donghai County in the Sulu UHP metamorphic belt of Eastern China, from the Chinese Continental Scientific Drilling Main hole. This study reports the results obtained from the analysis of oxide log data. A geochemical logging tool provides in situ, gamma ray spectroscopy measurements of major and trace elements in the borehole. Dry weight percent oxide concentration logs obtained for this study were SiO2, K2O, TiO2, H2O, CO2, Na2O, Fe2O3, FeO, CaO, MnO, MgO, P2O5 and Al2O3. Cross plot and Principal Component Analysis methods were applied for lithology characterization and mineralogy description respectively. Cross plot analysis allows lithological variations to be characterized. Principal Component Analysis shows that the oxide logs can be summarized by two components related to the feldspar and hydrous minerals. This study has shown that geochemical logging tool data is accurate and adequate to be tremendously useful in UHP metamorphic rocks analysis.

Joint Inversion of Geochemical Data and Geophysical Logs for Lithology Identification in CCSD Main Hole

The Chinese Continental Scientific Drilling (CCSD) main hole is located in the Sulu ultrahigh-pressure metamorphic (UHPM) belt, providing significant opportunities for studying the metamorphic strata structure, kinetics process and tectonic evolution. Lithology identification is the primary and crucial stage for above geoscientific researches. To release the burden of log analyst and improve the efficiency of lithology interpretation, many algorithms have been developed to automate the process of lithology prediction. While traditional statistical techniques, such as discriminant analysis and K-nearest neighbors classifier, are incompetent in extracting nonlinear features of metamorphic rocks from complex geophysical log data; artificial intelligence algorithms are capable of solving nonlinear problems, but most of the algorithms suffer from tuning parameters to be global optimum to establish model rather than local optimum, and also encounter challenges in making the balance between training accuracy and generalization ability. Optimization methods have been applied extensively in the inversion of reservoir parameters of sedimentary formations using well logs. However, it is difficult to obtain accurate solution from the logging response equations of optimization method because of the strong overlapping of nonstationary log signals when applied in metamorphic formations. As oxide contents of each kinds of metamorphic rocks are relatively less overlapping, this study explores an approach, set in a metamorphic formation model and using the Broyden Fletcher Goldfarb Shanno (BFGS) optimization algorithm to identify lithology from oxide data. We first incorporate 11 geophysical logs and lab-collected geochemical data of 47 core samples to construct oxide profile of CCSD main hole by using backwards stepwise multiple regression method, which eliminates irrelevant input logs step by step for higher statistical significance and accuracy. Then we establish oxide response equations in accordance with the metamorphic formation model and employ BFGS algorithm to minimize the objective function. Finally, we identify lithology according to the composition content which accounts for the largest proportion. The results show that lithology identified by the method of this paper is consistent with core description. Moreover, this method demonstrates the benefits of using oxide content as an adhesive to connect logging data with lithology, can make the metamorphic formation model more understandable and accurate, and avoid selecting complex formation model and building nonlinear logging response equations.

Performance of a modified particle swarm optimization in determining the thickness of the liquid layer outside a well casing pipe

The liquid layer outside a well casing develops from flaws along the cement–casing interface, which have been identified as leakage pathways in wellbores. The thickness of the liquid layer is one of the important parameters for determining cement bond quality. However, it is difficult to accurately determine the thickness of the liquid layer outside a well casing pipe with the traditional ultrasonic pulse echo method. To address this problem, we propose a novel inversion method for determining the thickness of the liquid layer between the casing and the cement sheath. According to test specimens, multi-layered structures are modeled to determine the overall reflected wave at the mud–casing interface. An improved particle swarm optimization algorithm is used to simultaneously inverse the thicknesses of the liquid layer, the casing, and the annular space between the casing and formation. Using synthetic data, this algorithm is more robust and stable when compared with standard particle swarm optimization inversion procedures. In addition, the influences of noise in the reflected wave and deviations of medium parameters on the inversion results of the liquid layer thickness are discussed. Finally, the method was used to determine the thicknesses of the liquid layer in test specimens. Numerical and experimental results demonstrate that the improved particle swarm optimization algorithm provides an effective approach to accurately determine the thickness of the liquid layer outside a well casing pipe.

Use of spectral gamma ray as a lithology guide for fault rocks: A case study from the Wenchuan Earthquake Fault Scientific Drilling project Borehole 4 (WFSD-4)

The main purpose of the Wenchuan Earthquake Fault Scientific drilling project (WFSD) was to produce an in-depth borehole into the Yingxiu-Beichuan (YBF) and Anxian-Guanxian faults in order to gain a much better understanding of the physical and chemical properties as well as the mechanical faulting involved. Five boreholes, namely WFSD-1, WFSD-2, WFSD-3P, WFSD-3 and WFSD-4, were drilled during the project entirety. This study, therefore, presents first-hand WFSD-4 data on the lithology (original rocks) and fault rocks that have been obtained from the WFSD project. In an attempt to determine the physical properties and the clay minerals of the lithology and fault rocks, this study analyzed the spectral gamma ray logs (Total gamma ray, Potassium, Thorium and Uranium) recorded in WFSD-4 borehole on the Northern segment of the YBF. The obtained results are presented as cross-plots and statistical multi log analysis. Both lithology and fault rocks show a variability of spectral gamma ray (SGR) logs responses and clay minerals. This study has shown the capabilities of the SGR logs for well-logging of earthquake faults and proves that SGR logs together with others logs in combination with drill hole core description is a useful method of lithology and fault rocks characterization.

A fast forward algorithm for real-time geosteering of azimuthal gamma-ray logging

Geosteering is an effective method to increase the reservoir drilling rate in horizontal wells. Based on the features of an azimuthal gamma-ray logging tool and strata spatial location, a fast forward calculation method of azimuthal gamma-ray logging is deduced by using the natural gamma ray distribution equation in formation. The response characteristics of azimuthal gamma-ray logging while drilling in the layered formation models with different thickness and position are simulated and summarized by using the method. The result indicates that the method calculates quickly, and when the tool nears a boundary, the method can be used to identify the boundary and determine the distance from the logging tool to the boundary in time. Additionally, the formation parameters of the algorithm in the field can be determined after a simple method is proposed based on the information of an offset well. Therefore, the forward method can be used for geosteering in the field. A field example validates that the forward method can be used to determine the distance from the azimuthal gamma-ray logging tool to the boundary for geosteering in real-time.

Integrated core-log interpretation of Wenchuan earthquake Fault Scientific Drilling project borehole 4 (WFSD-4)

Understanding slip behavior of active fault is a fundamental problem in earthquake investigations. Well logs and cores data provide direct information of physical properties of the fault zones at depth. The geological exploration of the Wenchuan earthquake Scientific Fault drilling project (WFSD) targeted the Yingxiu-Beichuan fault and the Guanxian Anxian fault, respectively. Five boreholes (WFSD-1, WFSD-2, WFSD-3P WFSD-3 and WFSD-4) were drilled and logged with geophysical tools developed for the use in petroleum industry. WFSD-1, WFSD-2 and WFSD-3 in situ logging data have been reported and investigated by geoscientists. Here we present for the first time, the integrated core-log studies in the Northern segment of Yingxiu-Beichuan fault (WFSD-4) thereby characterizing the physical properties of the lithologies(original rocks), fault rocks and the presumed slip zone associated with the Wenchuan earthquake. We also present results from the comparison of WFSD-4 to those obtained from WFSD-1, WFSD-3 and other drilling hole in active faults. This study show that integrated core-log study would help in understanding the slip behavior of active fault.

Crosswell electromagnetic modeling from impulsive source: Optimization strategy for dispersion suppression in convolutional perfectly matched layer

This study applied the finite-difference time-domain (FDTD) method to forward modeling of the low-frequency crosswell electromagnetic (EM) method. Specifically, we implemented impulse sources and convolutional perfectly matched layer (CPML). In the process to strengthen CPML, we observed that some dispersion was induced by the real stretch κ, together with an angular variation of the phase velocity of the transverse electric plane wave; the conclusion was that this dispersion was positively related to the real stretch and was little affected by grid interval. To suppress the dispersion in the CPML, we first derived the analytical solution for the radiation field of the magneto-dipole impulse source in the time domain. Then, a numerical simulation of CPML absorption with high-frequency pulses qualitatively amplified the dispersion laws through wave field snapshots. A numerical simulation using low-frequency pulses suggested an optimal parameter strategy for CPML from the established criteria. Based on its physical nature, the CPML method of simply warping space-time was predicted to be a promising approach to achieve ideal absorption, although it was still difficult to entirely remove the dispersion.

Machine Learning Interpretation of Conventional Well Logs in Crystalline Rocks

The identification of lithologies is a crucial task in continental scientific drilling research. In fact, in complex geological situations such as crystalline rocks, more complex nonlinear functional behaviors exist in well log interpretation/classification purposes; thus posing challenges in accurate identification of lithology using geophysical log data in the context of crystalline rocks. The aim of this work is to explore the capability of k-nearest neighbors classifier and to demonstrate its performance in comparison with other classifiers in the context of crystalline rocks. The results show that best classifier was neural network followed by support vector machine and k-nearest neighbors. These intelligence machine learning methods appear to be promising in recognizing lithology and can be a very useful tool to facilitate the task of geophysicists allowing them to quickly get the nature of all the geological units during exploration phase.