An HPO-Optimized LSTM Model for Prediction Problem of the Grasping Force via Surface Electromyography

Globally, there are approximately 3 million individuals with upper limb amputations, the vast majority of whom rely on prosthetics as assistive devices. However, current prosthetic hands face numerous limitations in controlling gripping force, which prompts us to explore more accurate prediction models. This paper studies the prediction problem of the grasping force by analyzing surface Electromyography (sEMG) signals for prosthetic hand control. An improved Long Short-Term Memory (LSTM) model whose parameters are optimized through the Hunter-Prey Optimization (HPO) algorithm is psesented. We then enhanced the model’s accuracy in predicting the transformation from electromyographic signals to force. Three assessment indicators, namely Mean Absolute Error (MAE), Mean Squared Error (MSE), and the coefficient of determination ( \(R^2\) ), are used to validate the model’s performance. The results demonstrate that the HPO-LSTM model outperforms the comparative models including Support Vector Machine (SVM) and Linear Regression (LR) models across all indicators, proving its potential in finely controlling the gripping force of prosthetic hands.