Abstract: Brain-machine interface (BMI) systems compensate for the lack of information due to damaged body parts via a combination of BMI technology and control theory. In this paper, we study a spontaneous single-joint motion task based on an experimental model of the psychoacoustic cortical neuron firing rate and the BMI control theory. In addition, we design a nonlinear decoder using an adaptive echo state network (ESN) and introduce the First Order Reduced and Contrdled Error learning (FORCE) algorithm to update the network's output weight. We verify the effectiveness of the designed decoder by simulating the performance of the decoder in the presence of natural ontological feedback information. Finally, using a direct inverse model framework based on a least-squares-support-vector-machine genetic algorithm, we design the optimal artificial sensory feedback of the neuronal firing rate in the sensory area of the cerebral cortex to stimulate the neurons of the cerebral cortex. The simulation results show that the designed closed-loop brain-computer interface (BMI) framework can restore the performance of a spontaneous single-joint natural motion task, which provides a new way to study the performance of closed-loop systems based on object input and output data when the system model is unknown.