Containing data noise in unconventional-reservoir-performance forecasting

Many decline-curve analysis methods have emerged for forecasting the future performance of unconventional reservoirs. However, severe noise in field data, coupled with the low-frequency rate in monitoring/reporting and the unknown behavior of many completion and reservoir parameters, collectively present serious challenges in obtaining correct model parameters in many settings. ress the complexity of this multitude of issues, performance forecasting is approached in two steps. First, we attempt to circumvent the data noise and frequency issues with a global cumulative production profile for a group of wells exhibiting similar performance characters, leading to the estimation of global model parameters. Second, we compare error trends amongst all methods for a basis of selecting well groups. Finally, a simple rule-of-thumb is developed to get an estimate of the allowable time for extrapolating performance prediction within 10% error.