4种智能算法在相平衡数据拟合中的应用

Application of Four Intelligent Algorithms in Phase Equilibrium Data Fitting

针对Marquardt-Levenberg法应用于多元物系相平衡数据拟合时,模型参数剧增,初值难以设定的难题,将4种智能算法,即遗传算法、神经网络,退火算法及粒子群算法,应用于相平衡数据的拟合。以正丙醇(1)+乙腈(2)二元物系汽液相平衡数据的Wilson拟合和甲醇(1)+乙腈(2)+1-乙基-3-甲基咪唑四氟硼酸盐(3)三元物系汽液相平衡数据的NRTL拟合为例,系统讨论了4种算法在应用时的主要影响因素,并将所得结果进行了分析和比较。结果表明遗传算法和粒子群算法可以较好地解决初值难设的问题,并且给出了每种算法的适用范围和使用建议。

The Marquardt-Levenberg (ML) algorithm is the most commonly used algorithm for phase equilibrium data fitting. However, this algorithm belongs to the local optimization algorithm. When ML algorithm was applied to multi-component system phase equilibrium data fitting, it is difficult to find the appropriate initial values for the general thermodynamic researchers owing to model parameters increasing sharply. In this paper, four kinds of intelligent algorithms, namely, genetic algorithm, neural network, annealing algorithm and particle swarm algorithm were applied to fitting the vapor-liquid equilibrium data of n-propanol ( 1)+ acetonitrile (2) binary system by Wilson model and fitting the vapor-liquid equilibrium data of methanol (1)+acetonitrile (2)+l-ethyl-3-methylimidazolium tetrafluoroborate (3)([EMIM][ BF4 ]) ternary system by NRTL model, re-spectively. The mainly influencing factors of four algorithms on the phase equilibrium data fitting application were discussed. The results were also analyzed and compared. On the basis of above work, the scope of the application and the use of recommendations of each method were proposed.