CCS and Power Systems
Carbon Storage - Geologic Storage Technologies and Simulation and Risk Assessment
Statistical Analysis of CO2 Exposed Wells to Predict Long Term Leakage Through the Development of an Integrated Neural-Genetic-Algorithm
Project No: FE0009284
Researchers at University of Louisiana at Lafayette (ULL), and their partners from Missouri University of Science and Technology and Schlumberger Carbon Services, will use novel statistical-neural genetic algorithm methods to study the leakage risks for wells exposed to CO2. At a typical CO2 injection site, the CO2 plume could reach abandoned and active wells. There is an increased risk of CO2 leakage through these wells, particularly if the well construction quality is not well known. In many cases, construction information for older wells is limited or does not exist. The methodology and output of this study may be used to evaluate the leakage risks from existing wells at current and future CO2 storage sites, as well as older wells where all of the well construction data is not available by comparing similar well attributes and risks between similar types of wells.
This study will first develop a comprehensive database of the wells in the Texas Gulf Coast area using electronic records that exist in the Railroad Commission of Texas, in addition to hard copy and microfilm data of older wells that exist in other government agency and private files. The database will be fed into a novel hybrid model of a neural-genetic algorithm, being developed as part of this project, to perform risk analysis across the database, and identify wells that should be subjected to remedial action. Statistical analyses will be performed by the model on general well attributes such as classification, well type, well construction details and materials, location, geology, geomechanical properties, wireline log data, mechanical integrity test data, plugging and abandonment reports, CO2 exposure duration, and reported well integrity problems of the wells. Multivariate statistical techniques, such as factor, regression and cluster, shall be used to analyze the database. In addition, the model output will be verified by a combination of wireline logs, core sampling, and testing that includes X-ray and scanning electron microscopy, cement logging, pH, and fluids testing.