吉布斯-赫尔姆霍尔茨图形神经网络用于预测在无限稀释下聚合物溶液的活性系数
Edgar Ivan Sanchez Medina1, Sreekanth Kunchapu1, Kai Sundmacher1,2
1Chair for Process Systems Engineering, Otto-von-Guericke University, Universitätsplatz 2, Magdeburg 39106, Germany.
这项研究增强了对聚合物混合物特性的机器学习. 一个修改的吉布斯-赫尔姆霍尔茨图神经网络 (GH-GNN) 准确地预测活动系数,改进了聚合物溶液建模.
科学领域:
- 化学工程是化学工程的重要组成部分.
- 计算化学计算化学
- 材料科学 材料科学 材料科学
背景情况:
- 机器学习模型擅长预测纯组件属性,但对混合物属性进行探索较少.
- 在聚合物加工等行业中,混合物特性至关重要,但预测仍然具有挑战性.
- 现有的模型经常与聚合物溶液的复杂性作斗争.
研究的目的:
- 修改吉布斯-赫尔姆霍尔茨图神经网络 (GH-GNN) 模型,用于预测在聚合物溶液中无限稀释时基于重量的活性系数.
- 在数据稀缺的场景中,研究不同聚合物表示对模型性能的影响.
- 利用转移学习来提高预测准确性和减少聚合物-溶剂混合物的可变性.
主要方法:
- 基布斯-赫尔姆霍尔茨图神经网络 (GH-GNN) 模型的修改.
- 评估各种聚合物表示 (单体,重复单元,周期单元,小聚体).
- 使用来自小型系统的活动系数数据应用转移学习.
- 与UNIFAC-ZM和热-FV模型进行比较分析.
主要成果:
- 修改后的GH-GNN模型在数据稀缺条件下对聚合物表示的敏感性降低.
- 转移学习显著提高了模型准确性,降低了预测变化.
- 该模型在混合物插入和溶剂抽取任务中实现了0.15的平均绝对误差.
- 与已建立的聚合物溶液属性预测模型相比,GH-GNN方法显示出更高的性能.
结论:
- 修改后的GH-GNN模型提供了一种强大的数据驱动方法,用于预测聚合物溶液的热力学特性.
- 转移学习有效地克服了聚合物-溶剂混合物建模中的数据限制.
- 基于图形神经网络的模型为复杂混合物属性预测提供了一个有希望的未来方向.
更多相关视频
06:55Synthesis of Cyclic Polymers and Characterization of Their Diffusive Motion in the Melt State at the Single Molecule Level
Published on: September 26, 2016
11:44Using Cyclic Voltammetry, UV-Vis-NIR, and EPR Spectroelectrochemistry to Analyze Organic Compounds
Published on: October 18, 2018
相关概念视频
Thermodynamics: Activity Coefficient
The activity coefficient is a measure of the deviation from ideal behavior. When the ionic strength of the solution is minimal, the activity coefficient of an ionic species is close to unity, making...
Factors Affecting Activity Coefficient
The activity coefficient value for an ion is close to one when the solution has almost zero ionic strength, i.e., when the solution shows close to ideal behavior. As the ionic strength of the solution increases from 0 to 0.1 mol/L, a...
Electrolytes: van't Hoff Factor
The colligative properties of a solution depend only on the number, not on the identity, of solute species dissolved. The concentration terms in the equations for various colligative properties (freezing point depression, boiling point elevation, osmotic pressure) pertain to all solute species present in the solution. Nonelectrolytes dissolve physically without dissociation or any other accompanying process. Each molecule that dissolves yields one...
Acid–Base Equilibria: Activity-Based Definition of pH
In solutions of very low ionic strength—for example, pure water—the...
Extraction: Effects of pH
Chemical Equilibria: Redefining Equilibrium Constant
To calculate the equilibrium constants of solutions of moderately high ionic strength, one must account for the salt effect. This redefined...
