High-Precision Lumped-Parameter Thermal Modeling and CNN-LSTM-Based Temperature Prediction for Optical Interconnect Modules

The increasing power consumption and intensified thermal coupling in next-generation high-speed optical interconnect modules present severe thermal management challenges. To address these issues, we propose a high-precision lumped-parameter thermal modeling method for optical interconnect modules. Building upon the constructed lumped-parameter thermal model (LPTM), a real-time transient temperature prediction framework utilizing a hybrid convolutional neural network and long short-term memory (CNN-LSTM) architecture is developed. The LPTM achieves high-precision modeling while reducing computational complexity. Compared with the simulation results from COMSOL Multiphysics 6.3, the proposed LPTM model exhibits an average mean absolute error (MAE) of only 0.52 °C and an average root mean square error (RMSE) of only 0.69 °C for the transient temperatures of key components. Furthermore, the CNN-LSTM neural network, trained on the dataset generated by the LPTM, achieves an MAE of 0.34 °C and an RMSE of 0.78 °C on an independent transient temperature testset generated by COMSOL Multiphysics. The proposed methodology successfully achieves both high-precision modeling and real-time, accurate transient temperature prediction, providing a solid theoretical foundation for thermal management in next-generation optical interconnect systems.