Shear wave travel time estimation from petrophysical logs using ANFIS-PSO algorithm: A case study from Ab-Teymour Oilfield

Among petrophysical logs, shear wave velocity is known to provide more accurate results in terms of mechanical rock properties. However, due to higher costs associated with acquiring shear waves, it is not a common practice to acquire shear wave at all wells. As such, many efforts have been made to estimate the log from other petrophysical logs. In the present research, shear wave travel time (slowness) is estimated from other petrophysical logs acquired at a well drilled into Ab-Teymour Oilfield, southwestern Iran. For this purpose, first, outliers were identified, using Tukey's method, and omitted from the dataset. Then, Non-dominated Sorting Genetic Algorithm (NSGA-II) and Multi-Layer Perceptron (MLP) methodologies were utilized to select effective input logs. Continuing with the research, two hybrid algorithms, namely Adaptive Neuro-Fuzzy System combined with Particle Swarm Optimization (ANFIS-PSO) and Adaptive Neuro-Fuzzy System combined with Genetic Algorithm (ANFIS-GA), were used to estimate shear wave from the selected logs. The results of hybrid NSGA-II and MLP on pre-processed data showed that, with increasing the number of inputs, the model's MSE decreases. It should be noted that the decreasing rate has no significant differences if the input parameters exceed four. Accordingly, using the hybrid algorithms on 4 selected logs (i.e. compressional wave, density, porosity, and depth logs), shear wave travel time (DTSM) was estimated. The estimation results indicated superior performance of the ANFIS-PSO model over that of ANFIS-GA. Furthermore, low difference between corresponding estimation errors to training and testing phases of ANFIS-PSO indicated that, the model can provide adequate reliability and accuracy when it comes to shear wave estimation. A comparison between the raw data-based and pre-processed data-based ANFIS-PSO models highlighted the efficiency of the pre-processing stage in enhancing the model results. In order to undertake a performance evaluation of the proposed model, the proposed method was compared to empirical correlations and multivariate regression analysis (MVRA). The results indicated significant superiority of the model over other models. The proposed method in the present paper can provide even more reliable and accurate outputs in terms of shear wave across the field understudy provided larger sources of data can be incorporated into the method, so as to save the costs required to acquire shear data.