Treatment of Missing and Corrupted Data in Waterfloods using the Reservoir Simple Models

Missing and corrupted data are main issues in oil industry and they cost up to $60 billion annually (Nobkhat, 2009). The focus here is on missing and corrupted data in production and injection rates for waterfloods. There are several missing patterns: Arbitary, monotone, multivariate and modified multivariate. The most severe pattern is modified multivariate pattern which happens when there is a major problem with the measurement system. Treating missing data has two options; 1- Dropping missing values, and it is not recommended for more than 3% missing. 2- Imputing the data using the simple reservoir models; Albertoni (2003) Resistance Model. A decision tree was programmed in Matlab to process these scenarios. Two outliers approaches were suggested in this work, a sample dependant and a distance based approaches which depend on the distance of each point from the neighbour points. It is not recommended to work with data has more than 6% missing. The understanding of data behaviour plays a main role in which approach will be treat missing values. Two cases were tested 15% and 30% of missing with max resulted value of R2 .95 and min .75, a software tool was developed to process these scenarios