Qualitative Logging Identification Of Water-flooded Zones In F Block Based On Random Forest

Positive rhythm deposits and homogeneous rhythm deposits are dominant in F block reservoir.F block reservoir has entered the middle and high water-flooded stages.Difficulty in interpretation of water-flooded zones increases.The accuracy of qualitative identification of water-flooded zones is low only through log curve amplitude.In addition,there are fewer samples for different water-flooded grades in this area.The morphological characteristics of log curves can better describe the change of sedimentary rhythm of reservoirs.Meanwhile,random forest is one of the mathematic methods to effectively solve the problem of qualitative logging identification of water-flooded zones.Combining the characteristics of log curve amplitude and amplitude difference with log curve morphology,qualitative logging identification of water-flooded zones based on random forest is researched,which is of great significance to improve the accuracy of qualitative logging identification of water-flooded zones with obvious rhythmic characteristics of reservoirs.In this paper,146 oil zones,149 water-flooded zones and 152 water zones are determined based on the method of dynamic and static combination.Nine morphological characteristics of log curves are extracted.The amplitude,amplitude difference and morphological characteristics of log curves in F block reservoir are analyzed.Select the characteristics with strong ability to distinguish oil zones,water-flooded zones and water zones which include deep laterolog resistivity,shallow laterolog resistivity,resistivity of flushed zone,amplitude difference of deep laterolog resistivity and shallow laterolog resistivity,amplitude difference of deep laterolog resistivity and flushed zone resistivity,relative center of gravity of formation density curve,relative center of gravity of deep laterolog resistivity curve,ellipticity of deep lateral resistivity,saturation coefficient of deep laterolog resistivity.Comparing the prediction accuracy by repeated experiments,the inner product function and penalty coefficient of support vector machine are adjusted.A qualitative identification model of water-flooded zones,oil zones and water zones in F block based on support vector machine is established.The node splitting criterion,maximum eigenvalue number,maximum depth of decision tree and number of classifiers of decision tree in random forest are adjusted to maximize the generalization ability of the model while guaranteeing the upper limit of accuracy.Therefore,a qualitative logging identification model of water-flooded zones,oil zones and water zones in F block based on random forest is established.By comparing the accuracy and generalization ability,the random forest model is superior to the support vector machine model.Random forest models are arrayed,and the classification results are merged with the original features in the form of probability as the input of xgboost model.Further,a water-flooded grade classification model of F block based on improved random forest and xgboost is established.By comparing the prediction accuracy and stability with traditional random forest model,the improved model has more advantages.79 wells in the area are processed by using the established identification model of water-flooded zone and the classification model of water-flooded grade.Among them,23 water-flooded wells are interpreted.8 strong water-flooded zones,22 medium water-flooded zones,42 weak water-flooded zones and 170 oil zones are determined.The predicted results have a very high coincidence rate with 90.1%.It shows that the qualitative logging identification model of water-flooded zone established in this paper can be well applied to log interpretation and evaluation of water-flooded zone in F block.