X
Advance Search
  • JianHuan Liu, JiaJun Zou, HuiRu Lei, Hao Chen, JianYe Chen. 2027: Physics-informed long short-term memory network for forecasting laboratory fault instability and uncertainty analysis. Earth and Planetary Physics: 1-11. DOI: 10.26464/epp2027002
    Citation: JianHuan Liu, JiaJun Zou, HuiRu Lei, Hao Chen, JianYe Chen. 2027: Physics-informed long short-term memory network for forecasting laboratory fault instability and uncertainty analysis. Earth and Planetary Physics: 1-11. DOI: 10.26464/epp2027002
Open Access    

Physics-informed long short-term memory network for forecasting laboratory fault instability and uncertainty analysis

  • Reliable forecasting of fault shear stress evolution is critical for understanding fault instability and mitigating seismic hazards. However, most laboratory earthquake studies have focused on inferring the current fault states rather than forecasting future shear stress evolution. We developed a novel physics-informed long short-term memory (PI-LSTM) network for fault shear stress forecasting in laboratory seismic cycles. It considers the ultrasonic wave speed and spectral amplitude as inputs and embeds the elastic coupling relationship between the fault and host rock into the LSTM framework to ensure mechanical consistency. We compared the PI-LSTM with a purely data-driven LSTM using double-direct-shear (DDS) experimental data under four training data ratios (70%, 50%, 30%, and 10%) and quantified the forecasting uncertainty using Monte Carlo dropout (MC dropout). The results show that the PI-LSTM performed comparably in data-rich scenarios but significantly outperformed the baseline under data-limited conditions. The physical constraint reduced the solution space, suppressed non-physical predictions, and reduced uncertainty, as reflected by the narrower 95% confidence intervals (CIs) and lower differential entropy, which further confirmed the improved reliability of the physically constrained forecasts. Overall, this study transformed a black-box model into a physically interpretable forecaster and offers a promising pathway for transferring lab-trained models to natural fault settings where observational data are scarce.
  • loading

Catalog

    Turn off MathJax
    Article Contents

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return