Sensitivity analysis of effective factors for estimating formation pore pressure using a new method: the LSSVM-PSO algorithm

The drilling of hydrocarbon wells is a process in which the drilling team deals with the numerous challenges to access hydrocarbon resources. Understanding the formation pore pressures is important to develop a successful and comprehensive drilling plan that minimize cost and maximize safety. This study evaluates the performance of some empirical models for calculating pore pressure based on petrophysical variables as input parameters. This research also compares the estimated performance of empirical models, efficiency assessment, and limitations caused by the petrophysical. The model presented in this study uses LSSVM-PSO artificial intelligence optimized neural networks as powerful tools in solving complex problems to identify complex relationship between petrophysical input data and the actual measured pore pressure with a modular formation dynamic measurement. Among the proposed network models, LSSVM-PSO, the most accurate model from performance and metric error, is a candidate for sensitivity analysis evaluation on ۱۵ different classes categorized by type and number of petrophysical input data. The best predictive approach among the specified classes belongs to the classes in which gamma-ray log petrophysical data participated as input nodes. This study confirms the effect of gamma log data as an influential factor in estimating the formation pore pressure parameter using artificial intelligence sensitivity analysis to the parameters assigned to the input variables. As can be seen in the results, the amount of RMSE = ۱.۱۳۸۹۵ and R۲ = ۱.۰۰۰۰ for class -۱۵ and for the total data used, which compared to other classes, these error parameters are much higher.Researchers in future studies can evaluate the results of this study as an efficient mathematical model.

Keywords:

pore pressure , LSSVM-PSO , Feature selection , Sensitivity analysis , Hybrid Machine Learning