Prediction of Gas Hydrate Formation Condition by Data-Driven Modeling: Different Machine Learning Models with Vector Quantization and Cuckoo Search Algorithm

Greenhouse gases can be defined as air pollutants that cause global climate warming. In order to reduce their harmful effects, these gases like methane and carbon dioxide can be stored in the form of compact gas hydrates. Prediction of gas hydrate formation conditions is very important for gas hydrate production and storage in industries. The goal of this study is to develop machine learning methods based on support vector regression and adaptive boosting models for predicting gas hydrate formation conditions for CO2 and natural gas. In this regard, SVR, AdaBoost.R2, VQ-SVR, VQ-AdaBoost.R2, CS-VQ-SVR, and CS-VQ-AdaBoost.R2 models have been developed and compared to obtain a model with the best performance. The cuckoo search optimization algorithm and vector quantization technique have also been utilized to determine the optimal values of the models’ hyper-parameters, reduce the computation time, and improve the accuracy and robustness of the models. As a result, since the values of the coefficient of determination and root mean square error for the CS-VQ-SVR model are 0.0215 and 0.9995, respectively, and the best agreement between predicted and actual values in this model’s graphs is obtained, it can be concluded that the CS-VQ-SVR model has the best accuracy and robustness among other developed models in predicting gas hydrate formation pressure with time. These results show that machine learning is viable for predicting the conditions of gas hydrate formation and preventing greenhouse gas emissions in industries.