Continuous prediction of human knee joint angle using a sparrow search algorithm optimized random forest model based on sEMG signals

Continuous prediction of human joint angles is crucial for enhancing the performance of man-machine cooperative control. This study proposed a random forest (RF) model optimized by the sparrow search algorithm (SSA) for achieving continuous motion prediction of the lower limb knee joint based on surface electromyography(sEMG) under various motion modes. The sEMG signal and knee angle were recorded during four human motions, namely normal gait, sitting-standing transition, ascending stairs and descending stairs. These signals were processed to remove noise and extract eigenvalues in both time and frequency domains. A knee angle prediction model based on SSA-RF algorithm was trained and evaluated using feature sample data. In the four motion mode experiments, the SSA-RF model achieved a minimum root-mean-square error of 1.569° for predicting knee joint angle, the average absolute error was only 1.05 °, and the coefficient of determination was as high as 0.99. The performance of the proposed model was compared with those of traditional backpropagation neural network, support vector machine regression, and random forest models. The comparison results clearly indicate that the proposed model can more effectively predict knee joint angles under various motion modes and performs better in promoting human-machine cooperation.