Deterministic Smart Tools to Predict Recovery Factor Performance of Saline Water Injection in Carbonated Reservoirs

Abstract

Throughout decades due to the scarcity of petroleum sources and weak performance of traditional waterflooding on increasing oil recovery factor (RF), enhanced oil recovery (EOR) processes have been applied to improve the ultimate oil recovery (Maghsoudian et al., 2020b). Hence, determining a suitable and cost-effective method to enhanced the ultimate recovery is still notable for developing oil fields, particularly in carbonated fields (Derkani et al., 2018). Among all EOR methods the role of low water salinity gets higher attention in numerous studies on both sandstone and carbonate reservoirs due to their potential advantages such as cost-effectiveness, simple preparation procedure, and appropriate stability (Kondori et al., 2020; Maghsoudian et al., 2020a). Low salinity (Losal) waterflooding prepared by diluting high salinity water containing various divalent and monovalent ions. According to previous studies, Losal has an immense impact on underlying mechanisms in petroleum reservoirs in order to change the wettability condition into water-wet state and dwindle the final residual oil (Liu and Wang, 2020). Besides, a large number of coreflooding tests analysis illustrated Losal has a great effect in both the secondary and the tertiary oil recovery process (Katende and Sagala, 2019). Experimental procedures commonly are time-consuming, high cost, and low accessibility. Therefore, applying artificial intelligence (AI) techniques as an alternative method to overcome the aforementioned barriers will be a suitable and trustworthy approach to predict objective function(s) and improve future practical research, in the absence of deep knowledge and related mathematical formulation to targeted procedures. The most well-known models based on AI are the artificial neural network (ANN), genetic algorithm (GA), neuro-fuzzy inference system (ANFIS), genetic programming (GP), least-squares support vector machine (LSSVM), etc (Li et al., 2020). These models have a great capability to develop precise and reliable output variables and have a low cost and swift computational procedure with high accuracy (Kondori et al., 2020). Despite numerous comprehensive research and data gathering in sandstone reservoirs, lack of sufficient comprehensive studies in carbonated reservoirs is still required due to the presence of harsh reservoir conditions such as heterogeneity, high temperature, and different physio-chemicals phenomenon (Hao et al., 2019). Based on the previous researches, precise studies on the effect of Losal waterflooding on recovery factor performance in carbonated reservoir by applying smart predictive models is still required. Thus, this paper purposed to cover this important gap by using practical deterministic models. The main purpose of this research is to introduce smart predictive tools such as ANN and multigene genetic programming (MGGP) for obtaining RF of LSWI process based on different parameters including porosity, permeability, temperature, injection rate, total dissolved salinity (TDS), viscosity and initial water saturation (Swi). These models are trained using 145 data point related to carbonated reservoirs. In the final step, the accuracy of the models are investigated based on statistical parameters such as root mean square error (RMSE), average relative error (ARE), and coefficient of determination (R2).