| Hangjinqi area in northern Ordos is a typical tight sandstone reservoir with rich oil and gas content.A large amount of well seismic data has been accumulated in the early exploration and development process,but the collected data exist shear wave velocity,the original fitting model applicability of shear wave velocity is poor,low accuracy of parts of seismic wave impedance inversion results limit.The accuracy of reservoir parameter calculation model needs to be further improved.In order to solve the problems mentioned above,this paper uses machine learning algorithm to predict reservoir parameters in the work area,and improves the accuracy of shear wave velocity data and wave impedance inversion results by data-driven method in Hangjinqi working area.In this paper,a large number of literatures about shear wave velocity prediction,seismic wave impedance inversion and machine learning are investigated,and random forest algorithm and long and short term memory neural network are used for research.In shear wave velocity prediction,the use of random forest algorithm and shear wave speed prediction model is established,both short-term and long-term memory neural network using two shear wave Wells in work area data for training and prediction,and with the DEM rock physics model and the nearest neighbor algorithm,comparing the results results show that both short-term and long-term memory neural network algorithm accuracy is the highest;In the wave impedance inversion prediction,the seismic track data near the key exploration Wells Jin86,Jin98 and Jin112 in the work area are modeled,and the wave impedance prediction is studied by using long and short-term memory neural network under supervised learning mode.The results of the inversion prediction are evaluated by blind well verification and comparison with the inversion results of conventional commercial software.The results show that the neural network algorithm can describe reservoir distribution more reasonably.The study area covers an area of about 31Km2,with 3d post-stack seismic records and well data from 3 key exploration Wells and 10 production Wells with incomplete data.Using data modeling in this area have been the prediction of shear wave velocity and wave impedance prediction error is low,can well with the measured data or earthquake information matching,and use the LSTM algorithm to study the shear wave velocity and wave impedance inversion data added data meet the engineering sense of reservoir parameters interpretation accuracy requirement. |
