In this paper, a human lower limb gait information prediction method based on a fusion of convolutional neural network (CNN) and long short-term memory network (LSTM) is proposed to solve the problem that lower limb rehabilitation exoskeleton cannot smoothly and accurately follow the movement of human lower limbs during the training process. The method first uses CNN to process the spatial features of human gait data, and extracts local features with spatial relationships through convolution and pooling operations. Then, the output of the CNN is passed to the LSTM for temporal dependency processing to capture the sequential pattern and evolutionary law of the motion for the prediction of the next gait state. In the experiments, a CNN-LSTM neural network model was built using Matlab to construct a dataset with the collected lower limb posture data, and the gait categories were used as the output for validation. The experimental results show that the method performs well in gait prediction assessment, and all predicted trajectories are strongly correlated with the measured trajectories with \( R^2\) greater than 0.999. Future research could further optimize the structure and parameter settings of the neural network model.