Improved neural network model based on dung beetle algorithm to predict CO2-brine interfacial tension

Abstract

Geological carbon sequestration refers to the permanent storage of captured CO2 through injection into subterranean saline or rock formations. The CO2-brine interfacial tension (IFT) is a crucial factor that significantly impacts the process's efficacy. Since the experimental determination of the IFT of brine and CO2 is both time-consuming and expensive, and a variety of sources of error may occur, developing a well-prepared and dependable model of CO2-brine IFT is crucial. In this paper, an attempt has been made to investigate the dung beetle optimization algorithm based back propagation neural network (DBO-BPNN) model for predicting CO2-brine IFT. The model contains 2616 experimental data of CO2-brine/water interfacial tension collected from various literatures, which can be divided into three regimes to be investigated: pure CO2-brine, pure CO2-water and impure CO2-water, and takes into account six independent variables: pressure, temperature, monovalent cation molality (Na+ and K+), bivalent cation molality (Ca2+ and Mg2+) in brine and the molar fractions of N2 and CH4 in the injected CO2 stream. The model's efficacy is assessed using a range of statistical and graphical techniques, and the model's validity is validated through the implementation of leverage methods, which identify anomalies across the entire dataset. Finally, the model is further compared with other intelligent models such as particle swarm optimization (PSO)-BPNN and grey wolf optimizer (GWO)-BPNN in terms of runtime, storage space and accuracy. According to the results, the DBO-BPNN model provides the best levels of accuracy and precision, with determination coefficient (R2), root mean square error (RMSE) and average absolute relative deviation (AARD%) of 0.9743, 1.598 and 3.16, respectively, and the R2 is enhanced by 0.8% and 2.2% in comparison to GWO-BPNN and PSO-BPNN models. Additionally, the DBO-BPNN model exhibits the least execution time, a reduction of 6.4% and 13.1% in comparison to GWO-BPNN and PSO-BPNN models, respectively. In addition, the DBO-BPNN model occupies storage space in the middle of the GWO-BPNN and PSO-BPNN models. The findings establish a dependable and robust framework that enables precise forecasting of the CO2-brine IFT.