Eliminating Subjectivity in History Match Evaluation Through Systematic Data Mining and Application of History Match Quality Index

Abstract

Abstract History matching is a process in which one investigates the most uncertain parameters of a reservoir model and adjusts them to achieve the best possible match with observed data. Although there exist standard processes for history matching, from manual methods to computer-assisted workflows, reservoir engineers with all levels of experience continue to find the assessment of matching quality a challenge. This paper compares three methods of assessing the quality of a history match: expert opinion, the Sum of Square of Error (SSE), and use of a History Match Quality Index (HMQI). Expert opinion, which is based on detailed visual inspection of modeled results, provides a sufficient assessment. However, it is often subjective and depends heavily on the reviewer's experience. The widely used weighted Root Mean Square (RMS) error (frequently used as an equivalent of SSE), which measures the degree of mismatch, is often affected by data outliers. Thus, it may not be practical in special cases. The HMQI method first assigns the match quality index a value of either 0 or 1, to classify the quality of match for each data point. The history matching quality index for each simulated parameter of a well is defined as the weighted average of all points. In addition, one can use Moving Linear Regression Analysis (MLRA) to analyze observed data. The MLRA identifies data trends and automatically removes outliers, depending on the confidence level used in the analysis. Then one uses the confidence interval obtained from MLRA to estimate the tolerance level in ranking the match quality index. We demonstrate the application of these methods and compare their relative merits in assessing the result of history matching in a mature oil field. MLRA clearly proves its ability to identify data trends and outliers. Combining MLRA and HQMI yields a superior and more robust quantitative result, compared with the other two most common methods: expert opinion and RMS error.