Abstract
Prediction of gas hydrate thermodynamic conditions is an important and attractive subject for its invaluable advantages. However, proposing a trustworthy and acceptable model to be applied for diverse cases is still rare. Existing models are limited to a definite group of materials. Hence, present study is accomplished for modeling various gas guest molecule hydrates. A novel hybrid fuzzy thermodynamic model is developed for predicting equilibrium gas hydrate formation temperature T (K) as a function of 1-guest molecule critical temperature Tc (K), 2-molecular weight Mw (g/mole), 3-acentric factor ω and 4-system pressure P (MPa). Sensitivity analysis is performed over naming parameters for respective importance. The categorized input parameters by optimized membership functions MFs (Fuzzifications) are verbally linked together to construct output verbal category (Fuzzy Inferences). Eventually the verbal output is transformed to a numeric value (Defuzzifications). 1-Research intuitions and inferences, 2-first principles of classic thermodynamics phase equilibrium, artificial neural network, ANOVA, historic experimental design and relative deviation analyses are all helped to construct and also optimize the hybrid fuzzy model via Mamdani and Sugeno FISs. Independent data of various guest molecules of refrigerants of R23, R13, R134a and R125a, hydrocarbons of CH(4), C(2)H(6) and other gases of CO, COS, N(2)O, H(2) and O(2) are assessed for prediction performance and cross validation. The selected materials have rarely been studied via an integrated model and the experimental data is few. Complex system of gas mixtures of CH4 + CO2 + H2S and CH4 + CO2 + N2 are successfully modeled via hybrid approach with AARD% of 0.579% and 0.774% respectively. The results show good agreement between the real and predicted values. The R-Squared equals to 0.9678 and regression line slope is 1.0252. The average absolute deviation AAD% of all data from hybrid fuzzy model equals to 0.299% (equivalent to 0.845 K) while 91.78% of data falling between - 1% (+ 2.826 K) and + 1% (-2.826 K) relative deviation. The statistical tools of diagnostic plots have successfully been passed by the model guaranteeing the safe predictive performance.