Predictive Analytics In Unconventional Resources

The oil and gas industry is no strangers to data.In the upstream sector,the advent of digitization and smart fields coupled with the boom in unconventional oil and gas production have led to a tremendous increase in volume of data generated in the industry in gigabytes,terabytes and petabyte(Big data).Unlike the conventional oil and gas field development and production,unconventional oil and gas production tend to defy the fundamental physical laws that govern multiphase flow of oil and gas in conventional fields.In addition,the capital intensive nature of unconventional field development,the multiple staged hydraulic fracturing and the overall economic output have spurred the innovations to economically develop and manage unconventional resources.One of such methods is leveraging the vast amount of data generated in unconventional field development and using big data analytics to build data driven models for field management purposes.This thus underscores the importance of analytics to the oil and gas industry,This thesis proposes machine learning algorithms(predictive analytics)to analyze the tight gas production data with the aim of building an analytic model that can forecast future gas performance leveraging the reservoir/fluid properties as well as the hydraulic fracturing parameters of the gas wells.K-means clustering will be used to cluster the fractured wells by initial gas production rate into two categories.The first category is a cluster of 5 groups(excellent,very good,good,average and poor)and the second category is a cluster group of 3(good,average and poor).The various groups in each category represent the performance of the wells.Artificial Neural Network will be used to build the predictive model while a General Linear Model will be used for comparison analysis.The Mean Square Error(MSE)will be used to determine how well a model fits the data set and will serve as the criteria for selecting the best model.The well initial gas production rate will again be classified in the look-back analysis using Monte Carlo simulation and the results will be compared with the cluster analysis.