基于长短期记忆网络的解码器设计及闭环脑机接口系统构建

LSTM-based decoder design and closed-loop BMI system formulation

随着脑机接口技术的发展,该技术在残疾人肢体功能恢复等方面应用越来越广泛。首先,在简介经典单关节信息传输模型基础上,设计并训练基于长短期记忆网络的解码器,代替原有脊椎电路通路将大脑信号传递给假肢;其次,为了在感觉反馈通路缺失时,仍能准确地恢复肢体运动功能,结合基于无模型控制策略设计的辅助控制器,构建闭环脑机接口系统,实现恢复关节活动障碍者缺失的感觉反馈通路从而实现跟踪期望轨迹的目的。由仿真可知,基于长短期记忆网络设计的解码器的离线解码效果良好。构建的闭环脑机接口系统对期望轨迹的跟踪以及缺失信息通路的恢复的结果验证了无模型控制辅助控制器良好的控制性能以及构建的闭环脑机接口系统的有效性。

With the development of brain-machine interface technology,the technology has been widely used in the restoration of limb motor function of disabled people.Firstly,referring to the classical single joint information transfer model,this paper designs and trains a decoder based on the long short-term memory network,which is used to replace the spinal circuit pathway to transmit the brain signal to the prosthesis.Secondly,in order to make up for the absence feedback channel and to restore limb movement function accurately,the designed decoder is combined with an auxiliary controller based on model-free control strategy to build a closed-loop brain-machine interface system which can restore the absence feedback path and achieve the purpose of tracking expected trajectory.Simulation results show that the decoder designed by the long short-term memory network has a good off-line decoding effect,that is,the designed decoder can well restore the absence spinal circuit channel.The results of tracking the desired trajectory and of restoring the absence information path of the constructed closed-loop brain-machine interface system verify the good control performance of the model-free control strategy and the effectiveness of the constructed closed-loop brain-machine interface system.