Long time prediction of human lower limb movement based on IPSO-BPNN

System delay caused by mechanical transmission, control calculation and data communication are the main factor affecting the man-machine collaborative control of lower extremity exoskeleton. Improved Particle Swarm Optimization Algorithm (IPSO) was proposed to optimize BPNN (Back Propagation Neural Network) to predict the future joint angle of human lower limb. The 3d motion capture system was used to collect the Angle data of human lower limb joints, and time span was added to reconstruct the time series, which was taken as the input of the model. Compared to PSO (Particle Swarm Optimization), IPSO added a three-route competitive optimization trajectory, the training feedback of BPNN and mutation operation, which accelerated the convergence of the algorithm and avoided local optimization. Besides, we established a prediction evaluation criterion with prediction duration, iteration efficiency, Root Mean Square Error (RMSE) and Determination Coefficient (DC) as the core to analyze the prediction results of BPNN, PSO-BPNN (Support Back Propagation Neural Network by Particle Swarm Optimization) and IPSO-BPNN (Support Back Propagation Neural Network by Improved Particle Swarm Optimization). The results show that the average RMSE of IPSO-BPNN is less than 0.75 and DC is more than 98%. IPSO-BPNN can make more accurate prediction of human lower limb joint angle, which is beneficial to improve the man-machine coordination performance of exoskeleton.