A hybrid ANN-GA method for analysis of geotechnical parameters

Due to the fact that the smallest principal field stress from pumping pressure graphs in leak-off tests (LOTs) by using traditional theory is always greater than the real values, this paper introduce a new method to improve estimation of the smallest principal field stress of a subsurface strata based on the information provided in LOTs, and the maximum horizontal in-situ stress is also improved. First, a geomechanical model on borehole pressurization is built to describe pressure changes in the borehole as well as borehole behavior during pressurization. Second, an artificial neural network (ANN) will be trained to map the relationship among the field stresses and borehole pressures. The cumbersomeness of building geomechanical model repeatedly can be thus avoided. Third, the genetic algorithm (GA) is used to characterize the field stresses of a strata with given borehole pressure information from a LOT. The results of the investigation demonstrate that it is feasible to analyze the principal field stresses based on LOTs data, using sophisticated geomechanical modeling and soft computing.