Infill Well Location Selection Procedures in Lost Hills Using Machine Learning

Abstract

The diatomite reservoirs in Lost Hills field provide an interesting reservoir management challenge due to its unique natures such as low permeability (~ 1 md) but high porosity (~ 50%), weak rock strength, and strong imbibition. In most conventional reservoirs, numerical simulation has been successful technology in forecasting, history matching and extraction of valuable information on optimum new well locations placement. However, numerical simulation results of Lost Hills were not successful due to the unique characteristics of the diatomite reservoir noted above. Therefore, the current procedures of selecting new well locations are based on the surface patterns and static reservoir properties. Machine learning (ML) has recently been spotlighted since it does not require specific physical models but can provide good estimations if there is enough data. Features like permeability, oil saturation, and other common reservoir properties also impact on production in diatomite reservoir, but the contribution of each property is not clearly understood. Because of its characteristics of not requiring physical models, ML has strong applicability in Lost Hills of which exact hydrocarbon production mechanisms are not clearly understood. In this study, a new methodology using Machine Learning was introduced for infill well location selection. This approach led to identification of approximate 550 infill well candidate locations, which were recommended further to the asset development team for execution. The infill well location selection workflow consists of historic data analysis, drainage area estimation, ML model training, and candidate location estimation. Using K-means clustering and statistical means, the characteristics of good producers are analyzed. Then, the drainage area for each well is approximated using Voronoi cells. This allows a relative good estimation of the remaining oil in place and non-draining areas. A support vector machine (SVM) model was trained to rank the locations based on their production capabilities. The new workflow provides more systematic means on new well location selection and enable the asset team to make data driven decisions.