One-dimensional convolutional neural network-based analysis of dielectric spectral abundance in liquid-phase media.

Aiming at the difficult problem of component information analysis of mixed dielectric spectra, the component information characteristics of mixed dielectric spectra are investigated by one-dimensional convolutional neural network, and the component analysis of mixed media is realised. First, the mixed dielectric spectra of water, ethanol and isopropoxyethano (iso) with different volume ratios were obtained by experimental measurements, and the singular spectrum analysis (SSA) method was applied to denoise the raw data, which provides a new data processing method for the effective analysis of dielectric spectra. Then, Utilizing the linear mixing model, we systematically obtained the dielectric spectra of binary mixtures of pure water and isopropoxyethanol with diverse ratios, along with those of multi-component mixtures integrating water, ethanol, and isopropoxyethanol at various proportion settings. The generated data were used as a training set for a one-dimensional convolutional neural network to model the correlation between the mixed dielectric spectra and the mixing ratios. The coefficient of determination (R²) values of the model for the two-component and three-component mixed solution test sets were both found to be 0.999. In the validation sets, the corresponding R² values were determined to be 0.9887 and 0.9786, indicating that the accuracy and reliability of the unmixing algorithm based on the one-dimensional convolutional neural network in predicting the dielectric spectra of different proportions of the mixed media are good. This study provides an effective method to analyse and predict the dielectric properties of mixed media integrated apparent dielectric spectra, which provides a scientific basis for the research and analysis of the dielectric properties of mixed media, in addition to this study, this study promotes the integration of dielectric spectroscopy analysis and machine learning, providing new ideas and tools for the research of water environment governance and hydrochemical analysis.