Deriving activity coefficients from partial excess Raman spectra

混合物的过剩性质是指在相同的温度、压力和组成下，真实的和理想的混合物的性质之间的差异。根据“最新技术水平”，可以通过偏最小二乘回归（PLSR）从同一混合物的过量吸收光谱中提取混合物的过量吉布斯能。混合物过剩性质的相关性有两个缺点：（1）它们是混合物化合物的函数。因此，各种“族”的混合物，如酮/烷烃或酮/醇，不能用同一个相关性来反映。(ii)通过将gE模型拟合到由相关性产生的gE值，可以获得期望的活度系数。 如果考虑混合物的部分过剩性质而不是混合物的过剩性质，这些缺点就可以避免。因此，本项目旨在将混合物的部分过量拉曼光谱与其活度系数进行关联。提出的混合物化合物的部分过量性质的关联具有两个优点相比，“国家的最先进的”：（i）作为部分过量的性质被认为是，所确定的相关性应该是更独立的混合物的家庭。与混合物族的独立程度取决于相关性方法，因为将使用机器学习方法，该方法可以考虑线性（偏最小二乘回归）以及非线性（卷积神经网络）关系。(ii)所需的活性系数直接从所确定的相关性的结果，并不一定要通过拟合第一gE模型的相互作用参数获得。部分过量性质的确定的相关性将使实验拉曼光谱测定的化合物在混合物中的活度系数，而不需要测量相平衡。

Excess properties of mixtures quantify the difference of the property of a real and an ideal mixture at identical temperature, pressure and composition. According to the “state-of-the-art” the excess Gibbs energy of a mixture can be extracted via partial least squares regression (PLSR) from the excess absorption spectrum of the same mixture. The correlations of excess properties of mixtures feature two disadvantages: (i) they are a function of the mixture compounds. Thus, mixtures of various “families”, such as ketone/alkane or ketone/alcohol, cannot be reflected with one and the same correlation. (ii) The desired activity coefficients can be obtained having fitted gE models to the gE values that resulted from the correlations. These disadvantages can be circumvented, if partial excess properties of the mixture compounds are regarded instead of excess properties of mixtures. Therefore, this project aims at correlating the partial excess Raman spectrum of the mixture compounds with their activity coefficients. The proposed correlation of partial excess properties of mixture compounds features two advantages compared to the “state-of-the-art”: (i) As partial excess properties are regarded, the identified correlations are supposed to be more independent from the mixture family. The degree of independence from the mixture family depends on the correlation method, because of which machine learning methods will be made use of that can regard linear (partial least squares regression) as well as non-linear (convolutional neural networks) relations. (ii) The desired activity coefficients result directly from the identified correlations and do not have to be obtained by fitting first the interaction parameters of gE models. The identified correlations of partial excess properties will enable the experimental Raman spectroscopic determination of activity coefficients of the compounds in a mixture without the need of measuring phase equilibria.