Abstract: Accurate prediction of reservoir physical parameters is crucial for detailed reservoir characterization. Deep learning technology, due to its powerful data structure mining capabilities, has been extensively and deeply studied in this field, but current research mainly focuses on the application of neural networks such as LSTM and DNN. Liquid time-constant networks (LTC), as an emerging neural network architecture, has shown great potential in areas such as time series forecasting, utilizing dynamic reservoirs to process time series data. Unlike traditional feedforward neural networks, LTC includes a dynamic "liquid" layer, enabling it to capture and utilize complex temporal dependencies in the input sequence. Therefore, this paper proposes a prediction method based on Liquid time-constant networks (LTC) for accurate prediction of reservoir physical parameters. This model combines a one-dimensional convolutional neural network (CNN) with a bidirectional liquid time-constant networks, using elastic parameters (P-wave, S-wave velocity, and density) as input to predict porosity, water saturation, and clay content. LTC, through its adaptive time constant, can effectively capture the complex temporal dynamics and non-stationary characteristics of well logging data. Experimental results show that, compared with traditional fully connected neural networks, this LTC-based prediction method reduces the root mean square error of predictions for porosity, water saturation, and clay content by 62.3%, 10.3%, and 30.4%, respectively. This study confirms the significant advantages of liquid time-constant networks in processing geophysical data with complex temporal characteristics, providing a more reliable new technical approach for fine reservoir characterization.