Abstract: Accurate prediction of reservoir petrophysical parameters is of great significance for detailed reservoir characterization. However, existing methods struggle to effectively model multi-scale complex temporal features. To address this, this study proposes a reservoir petrophysical parameter prediction method using liquid neural networks (LNN). This method employs a cascaded architecture: first, a one-dimensional convolutional neural network (CNN) extracts local geological features from raw elastic parameters (such as P-wave velocity, S-wave velocity, and density); subsequently, the extracted feature sequence is fed into a bidirectional liquid neural network (Bi-LNN), which leverages its unique liquid time-constant mechanism to achieve precise mapping from elastic parameters to porosity, water saturation, and clay content. Experimental results demonstrate that the LNN model exhibits superior representation capability and adaptability to the multi-scale, non-stationary temporal features inherent in log data. Compared to traditional fully connected deep neural networks (DNNs), the proposed LNN-based method leverages its temporal modeling advantages to significantly enhance prediction accuracy, showing higher precision in predicting porosity, water saturation, and clay content, and thus providing a more reliable approach for detailed characterization of reservoir petrophysical parameters.