LS-SVM Combined with ZNN for Predicting the Continuous Motion Joint Angle of Lower Limb

In this paper, a method fusing least squares support vector machine (LS-SVM) with Gaussian kernel function and zeroing neural network (ZNN) is proposed to forecast the continuous motion of lower limb. The surface electromyography (sEMG) signal contains human behavior information and directly reflects the movement intention. In the experiment, the sEMG signal of the lower limbs when the tester is doing leg extension exercise is collected and the real knee joint angle is recorded at the same time. Then the raw sEMG is subjected to a series of preprocessing, and the corresponding muscle activation is gained by calculation. The muscle activation is used as input to the LS-SVM model, and the output to the model is the knee joint angle. LS-SVM transforms the original problem into solving linear equations. When the amount of data is relatively large, the traditional solution method is very time-consuming, and the proposed ZNN method is able to solve the problem, thus speeding up the convergence speed and greatly reducing the learning time. Finally, the back propagation neural network (BPNN) model is utilized to form a comparative experiment. The numerical results indicate that a more stable and better performance is reflected in the raised method, which provides a valuable reference for the research of joint continuous estimation.