Prediction of Shear Wave Velocity by Extreme Learning Machine Technique from Well Log Data

Shear wave velocity (Vs) is one of the key geomechanical parameters effective in the drilling of hydrocarbon reservoirs. In this study, a novel machine learning (extra learning machine (ELM)) approach is developed to predict Vs based on four input variables obtained from well log, including neutron porosity (NPHI), bulk density (RHOB) and gamma-ray (GR). Two algorithms multi-layer perceptron (MLP) and ELM and various empirical equations (Brocher, Eskandari et al., Castagna et al. and Pickett) have been used to predict Vs in this paper. The results show that the performance accuracy for these models includes: ELM> MLP> Castagna et al. > Eskandari et al. > Pickett> Brocher. So, the result that shows the ELM model has higher accuracy than the other machine learning (MLP) approach and also other empirical equations (RMSE = 0.0444 km/s and R2 = 0.9809). Some advantages to the other artificial neural network approach include higher accuracy and performance characteristics, simple algorithm learning, improved performance, nonlinear conversion during training, no stuck in local optimal points, and it is over fitting. The novelty used in this paper is the type of newly implemented artificial model (ELM) and the number of input parameter. This approach possesses to the higher power, speed and accuracy than the methods used by other researchers to predict Vs.