RIME-Optimized CEEMDAN-LSTM Framework for sEMG Prediction: Cross-Subjects, Tasks, and Ages Study

Surface electromyography (sEMG) signals are essential for monitoring muscle activation which are widely used in various fields, such as rehabilitation robotics and human-computer interaction systems. However, it remains a challenge in accurately predicting and analysing sEMG signals due to their nonlinearity, non-stationarity and individual differences. Therefore, this study proposed a novel prediction model named RCEEMDAN-RLSTM to address these limitations This model integrated an adaptive noise reduction technique with a feature extraction strategy, introducing Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) for signal decomposition, coupled with a Long Short-Term Memory (LSTM) network for temporal pattern learning. Additionally, the Rime Ice Optimization Algorithm (RIME) was employed to optimize the parameters of both components in real-time. The RIME-based optimization algorithm enables dynamic adjustment of model parameters, enhances noise suppression and accommodating individual differences in sEMG signal characteristics. To evaluate the robustness of the proposed model, we computed the Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Prediction Accuracy (PA) using experimental sEMG data collected from diverse muscle groups, motor tasks, and age groups in 13 healthy subjects. Quantitative analysis revealed statistically significant improvements across all evaluation metrics, achieving an average 48.42% reduction in MAE and 56.32% ± 18.75% decrease in RMSE across muscles. The prediction accuracy rate achieved 97.59%. The multi-type sEMG-driven approach also achieved 98.91% stability. These findings validate the effectiveness and reliability of the RCEEMDAN-RLSTM method in estimating sEMG signals. This has important implications for improving rehabilitation control strategies and advancing human-machine interaction systems.