Research And Application Of Training Image In Reservoir Multi-point Statistical Modeling

Training images are basic tool of multi-point geo-statistics, introducing geological priorknowledge or conceptual models into reservoir modeling through training images is abreakthrough contribution of multiple-point geo-statistics to reservoir modeling. A trainingimage is a numerical prior geological model which contains the facies structures and spatialdistribution patterns which are believed existing in realistic reservoirs.It is a combination ofvarious types of data and information (well logging and seismic).This paper studies the principle and method of establishing training images, buildingfour different training images according to Venezuelan M blocks’ geological features, thenbuilds four different training images’ facies models of Orinoco heavy oil belt M Reservoir bymulti-point geo-statistical method. When using the logging data modeling，the simulatedresults of four training image can be compared, selecting training image according tosimulation results in line with the geological conditions. When using the seismic dataconstraint modeling, simulation results of one set training images were analyzed, thencompares the simulation results of four different training images.We find that training image has a great influence on simulation results when only usingthe logging data. In the case of seismic constrained reservoir modeling, even using differenttraining image, the simulation results can be much the same.