Scuola di Architettura e Design, Università di Camerino, 63100 Ascoli Piceno, Italy.
Departamento de Física Aplicada, Universidad de Extremadura, 06006 Badajoz, Spain.
Molecules. 2021 Mar 15;26(6):1636. doi: 10.3390/molecules26061636.
An artificial neural network model is proposed for the surface tension of liquid organic fatty acids covering a wide temperature range. A set of 2051 data collected for 98 acids (including carboxylic, aliphatic, and polyfunctional) was considered for the training, testing, and prediction of the resulting network model. Different architectures were explored, with the final choice giving the best results, in which the input layer has the reduced temperature (temperature divided by the critical point temperature), boiling temperature, and acentric factor as an independent variable, a 41-neuron hidden layer, and an output layer consisting of one neuron. The overall absolute percentage deviation is 1.33%, and the maximum percentage deviation is 14.53%. These results constitute a major improvement over the accuracy obtained using corresponding-states correlations from the literature.
提出了一种用于涵盖宽温度范围的液体有机脂肪酸表面张力的人工神经网络模型。为了训练、测试和预测所得网络模型，考虑了 98 种酸（包括羧酸、脂肪族和多功能酸）的 2051 组数据。探索了不同的架构，最终选择了给出最佳结果的架构，其中输入层的独立变量为缩减温度（温度除以临界点温度）、沸点和偏心因子，隐藏层有 41 个神经元，输出层只有一个神经元。整体绝对百分比偏差为 1.33%，最大百分比偏差为 14.53%。与文献中相应状态相关性获得的精度相比，这些结果是一个重大改进。
