The movement speed of the slave manipulator arm in traditional space teleoperation system completely depends on the operating speed of the operator. In order to improve the safety of the space teleoperation system, an adaptive speed control method based on the recognition of the operating speed of the operator is proposed. Combining with the theory of deep learning, a fusion model based on convolutional neural network (CNN) and gate recurrent unit (GRU) neural network is proposed to identify and classify the speed of operator. Nine subjects were selected to construct an operator speed sample library. The operating speed of the operators is divided into three categories, and the final recognition accuracy rate reaches 92. 71% . And, on this basis, the cascade PID is used to realize the adaptive speed control of the slave manipulator arm. Experiments confirm that the model can also accurately identify new operators. At the same time, the accuracy of the model is better than that of the fusion model of convolutional neural network and recurrent neural network (RNN), and the real-time performance of the model is better than that of the fusion model of convolutional neural network and long short-term memory (LSTM) neural network. Besides, the adaptive speed control based on this model can reduce the end linear speed of the manipulator arm while ensuring that the movement trajectory of the slave manipulator arm remains unchanged, which helps to improve the safety of the space teleoperation system.