An improved probability combination scheme based on principal component analysis and permanence of ratios model - An application to a fractured reservoir modeling, Ordos Basin

摘要

A two-step reservoir modeling process is preferred in the presence of multiple secondary data. In this method, multiple selected secondary data are first combined to generate one spatial probability distribution of primary data by probability combination schemes (PCS), and the intermediate result is then integrated with well data through spatial modeling. Probability combination of secondary data is the key for constructing accurate reservoir models. The traditional PCS assumes that secondary data are independent of each other. However, secondary data are in fact commonly redundant, and it is difficult to fully quantify the data redundancy and obtain the optimal adjustment weights. As a result, the estimated probability by the traditional method is often seriously biased. We propose an improved PCS based on principal component analysis (PCA) and permanence of ratios (PR) model in order to reduce the bias. First, a target-based PCA is developed to create a set of independent principal components that satisfy the assumption made in the PR model. Second, the elementary conditional probabilities of primary data are obtained by calibrating each selected principal component. Finally, the PR model is used to estimate the joint conditional probability of primary data by linking the elementary conditional probabilities from the second step. We illustrate the detailed workflow of the improved PCS by applying it to a fractured reservoir modeling case in the Ordos Basin. In this example, the probability estimated by the improved PCS had 41% less error than the traditional PCS. Furthermore, the established fracture models based on the improved PCS were validated to have a better match with the dynamic data, reducing the error of simulated water cut by 63%. The improved PCS can also be widely used for spatial modeling of all types of categorical or continuous variables.