Prediction of lower limb joint angle based on convolutional neural network(CNN)leads to the loss of partial human electromyography signal feature information

resulting in poor overall prediction.Aiming at this problem

a lower limb joint angle prediction method combining attention mechanism（AM）-based long short-term memory(LSTM)network and CNN model is proposed.The human surface electromyography signal and the lower limb joint angles captured by motion capture are collected

and muscle activity features mapped from muscle signals are used as the inputs to conduct research of the lower limb joint angle prediction.The research results show that in the angle prediction of hip

knee and ankle joint angles of lower limbs

the CNN-LSTM model based on AM reduces a mean absolute error(MAE)by 26.4% and 53% on average compared with the CNN-LSTM model and CNN model

respectively

and the root mean square error(RMSE)value is reduced by 32.3% and 50.9% on average.It is verified that the model has higher accuracy in the prediction of lower limb joint angle.