In this paper, the single leg joint of pneumatic muscle quadruped robot is taken as the control object. Based on the establishment of dynamic model and sliding mode controller, an online identification and compensation control strategy of model parameters based on error feedback is proposed. In order to enhance the anti-disturbance ability of the pneumatic muscle driving joint, an anti-interference control strategy integrating the muscle spindle reflection is proposed. Firstly, the dynamic model parameters are identified online by recursive least square method with forgetting factor, and the angle error is fed back to the input side for real-time update of the dynamic model parameters. Secondly, the change rate of pneumatic muscle contraction velocity under disturbance is analyzed, the muscle spindle reflection pathway is fused to compensate for disturbances. Finally, the simulation results show that the root-mean-square error of the dynamic parameter identification and compensation controller is reduced by 14.82% and 21.84% respectively under the two equivalent disturbances. and the mean absolute error is reduced by 47.27% and 56.25% respectively. After the spindle reflection system is fused, the expected signal under disturbance is compensated. The mean absolute error is reduced by 38.55% and 62.16%, and the root mean square error is reduced by 31.94% and 59.18%, respectively, when there is step interference and slope interference.