LLE data are important for simulation and design of extraction equipment. In this study, deep neural network (DNN) structure was proposed for modelling of the ternary liquid-liquid equilibrium (LLE). LLE data of (water + butyric acid + 5-methyl-2-hexanone) ternaries defined at three different temperatures of 298.2, 308.2, and 318.2 K and $P$ = 101.3 kPa, were obtained experimentally and then correlated with nonrandom two-liquid (NRTL) and universal quasi-chemical (UNIQUAC) models. The performance of the proposed DNN model was compared with that of NRTL and UNIQUAC in terms of the root mean square errors (RMSE). RMSE values were obtained between 0.02-0.06 for NRTL and UNIQUAC, respectively. For DNN, the error values were obtained between 0.00005-0.01 for all temperatures. According to the calculated RMSE values, it was shown that proposed DNN structure can be better choice for the modelling of LLE system. Othmer-Tobias and Hand correlations were also used for the experimental tie-lines. Distribution coefficient and separation factors were calculated from the experimental data.

LLE 数据对于提取设备的模拟和设计非常重要。在这项研究中，提出了深度神经网络 (DNN) 结构来模拟三元液液平衡 (LLE)。在 298.2、308.2 和 318.2 K 三个不同温度下定义的（水 + 丁酸 + 5-甲基-2-己酮）三元的 LLE 数据和$磷$= 101.3 kPa，是通过实验获得的，然后与非随机双液体 (NRTL) 和通用准化学 (UNIQUAC) 模型相关联。在均方根误差 (RMSE) 方面，将所提出的 DNN 模型的性能与 NRTL 和 UNIQUAC 的性能进行了比较。NRTL 和 UNIQUAC 的 RMSE 值分别在 0.02-0.06 之间。对于 DNN，所有温度下的误差值都在 0.00005-0.01 之间。根据计算出的 RMSE 值，表明所提出的 DNN 结构可以更好地选择 LLE 系统的建模。Othmer-Tobias 和 Hand 相关性也用于实验联系线。根据实验数据计算分配系数和分离因子。