Methods Research On Logging Lithology Identification For Intermediate/Basaltic Rocks In Liaohe Basin

With the increase of oil and gas resources demand,Volcanic rocks havebecome a important goal of the oil and gas exploration. The easterndepression in Liaohe Basin has geologic characteristic of multiphasevolcanic eruption and complicate lithofacies/lithology. The reservoir ismade up of intermediate and basaltic volcanic rock. Because of low rateof core in volcanic rock reservoir and development costs in explorationprocess，so a gain of core date is relatively small. Using a small numberof core date， Complete and systematic evaluation of the reservoir isdifficult. Compare with core data，Well-logging data is rich and full，continuous and strong，and has vertical accuracy，a more comprehensivereflection of the reservoir characteristics，so using well-logging data inreservoir evaluation，it seems particularly important. Due to the rockformed in a different way，complexity and grain-level diversity of mineralcomposition and other reasons，lithology identification of volcanic rock ismore difficult.Using well-logging data to ithology identification in recent years，mainly in the conventional logging crossplot technique， multivariatestatistical analysis，acoustic and electric imaging logging and ElementalCapture Spectroscopy (ECS)，these methods take attention to innovation oflogging methods and application of advanced logging instruments. On the basis of learning domestic and foreign advanced technologyand experience，according to the volcanic rocks characteristic in Liaoheoil field，this paper focus on well-logging combined with geological data，make full use of conventional well-logging data for quantitativeevaluation of volcanic reservoir based on core data analysis，and sum upthe suitable and operable logging method for volcanic rocks in Liaoheoilfield. Major research includes:(1) Summary of geological and logging response characteristicsfor Intermediate/Basaltic Rocks in Liaohe basin.The lithology of volcanic rocks in eastern depression of Liaohe basindominated by intermediate/basaltic rocks in currently drilling results，inaccordance with the principles of volcanic rocks classification，combined with actual demand of the volcanic reservoir exploration andthe feasibility of well-logging identification，we classify the volcanicrocks in the study area. To evaluate the accuracy of lithologyidentification and analyze the variation between lithology and eachparameter of log responses，we summarize the logging response ofintermediate/basaltic rocks in Liaohe basin， and extract loggingparameters including gamma ray (GR)、 compensated neutron logs(CNL)、acoustic (AC)、density (DEN)、Resistivity lateral log deep(RLLD)，which sensitive to lithology.(2) Research on methods of volcanic rock lithology identification based on binary and multiless support vector machine (SVM).First， the principle of binary and three classical multiclass SVM(‘one-against-rest’，‘one-against-one’ and ‘directed acyclic graph’) wereanalyzed; Second，cross validation and grid searching algorithm wereadopted to optimize the penalty factor and kernel parameter; Then，training samples were selected from logging data which core data areintegrity and conbentional logging curves are complete in108wells，andthree SVM lithology identification models were constructed with thesetraining samples; Lastly，corresponding sections of logging data in4testing wells were taken as testing data to lithology identification，and theidentification results were compared with core data.(3) Research on methods of volcanic rock lithology identificationbased on fractal dimension of logging curves.On the basis of fractal theory and the self-similarity of loggingcurves， box-counting， Correlation dimension and rescaled analysistechnique are adopted to calculation methods of box-counting dimension，correlation dimension and rescaled analysis dimension of logging curves.Through an analysis of logging curves from the volcanic reservoirs in theeastern depression， Liaohe basin， we calculate the box-countingdimension，correlation dimension and rescaled analysis dimension oflogging curves from108wells，965m sections， and discuss therelationship between fractal dimension of the three logging curves and the texture of volcanic rocks. To proof the accuracy of conclusion， wecalculate the box-counting dimension，correlation dimension and rescaledanalysis dimension from4testing wells and predict the texture ofvolcanic rock in certain depth，compare the predicting result with thelithology of core data.This paper summarizes the previous research work， and theinnovation lies in the following areas:We have applied directed acyclic graph strategy in multiclass SVMto volcanic rocks lithology identification fields.We have excavated fractal characteristics of logging curves andcalculate logging curves fractal dimension using calculating methods ofbox-counting dimension， correlation dimension and rescaled analysisdimension.We study the rock types of volcanic rocks from geological material.We find combination characteristics of logging curves from loggingmaterial. So the geological material and logging material are collated，weeventually reached the purposes of lithology identification.Through the above work，we get following awareness:(1) Geologic and logging response characteristics for volcanicrocks.Volcanic rocks with different rock types，corresponds to a certainlogging response characteristic combination. According to the log response features can determine the type of volcanic rocks，However，frequent secondary changes in the development of volcanic rocks canlead to the change of mineral and chemical composition of volcanicrocks. Later tectonic movements and dissolution，will make the volcanicrocks develop a lot of cracks and pores. Therefore，logging responsecharacteristics of volcanic rock will also change.(2) Lithology identification for SVM methodAlthough the classification algorithms of ‘one-against-rest’，‘one-against-one’ and ‘directed acyclic graph’ have different identifyingaccuracy， but we can determine thickness and boundary of differentlithology formations in Liaohe basin based on logging response features.The identifying results of4testing wells show that the accuracy of‘directed acyclic graph’ method is highest. SVM cannot distinguish thelithology that are similar in mineral and chemical composition， butdifferent in structure and texture. The sample spaces of binary andmulticlass SVM are established using conventional logging material. Inthe lack of imaging logging data，element logging data，and other speciallogging data，SVM method are still applicable.(3) Lithology identification for fractal dimension of logging curvesWell logging curve is a fractal system with self-similar structure，Which degree of complexity can be represented quantitatively by usingfractal dimension. The methods of box-counting dimension，correlation dimension and rescaled analysis can calculate the fractal dimension oflogging curves. When applying logging curves in fractal dimensioncalculation， the division number in box-counting dimension， thedetermination of non-scale interval in correlation dimension and thechoice of sample points，to a certain extent，affect the accuracyof thecalculation of fractal dimension. By calculating the fractal dimension thefive logging curves of GR、RLLD、AC、DEN and CNL，find the fractaldimension of different logging curves about one lithology are basicallysame; Using the difference of logging curve fractal dimension，we canmake the prediction of volcanic rocks textures. The texture of volcanicrock is more complex， and the fractal dimension are larger. Thebox-counting dimension of lava is smaller than the pyroclastic rock. Atpresent，the effect of application of fractal dimension to distinguish thesupergene rock from sub-volcano rock，or separate plutonic rock fromvolcanic lava is not obvious.