An Adaptive ADRC Control for Parkinson’s Patients Using Machine Learning

Parkinson's disease (PD) is one of the most common diseases that its main complications are hand and head tremors and inflexibility of muscles. One of the prevalent treatments that employ for reducing the symptoms of that is deep brain stimulation (DBS). In practice, a sensor is located in the patient's finger for detecting and evaluating the tremor values in PD. Using an open-loop control structure for stimulating one area of basal ganglia (BG) is the common approach, but in this work, two areas of BG, named subthalamic nucleus (STN) and globus pallidus internal (GPi) are stimulated in a closed-loop manner separately for i) reducing the intensity of electric field and consequently disappearing the side effects of DBS ii) decreasing hand tremor. In particular, an adaptive Active Disturbance Rejection Control (ADRC) based on a deep deterministic policy gradient (DDPG) and a conventional feedback controller are presented for simultaneous stimulating STN and GPi, respectively. In this way, the control coefficients of the ADRC are considered as the control objective parameters that are designed by the actor and critic neural networks (NNs) of DDPG. The suggested scheme is applied to a BG system model which is frequently studied in the literature. The comprehensive simulation studies are accomplished to confirm the supremacy of the ADRC based DDPG scheme over the state-of-the-art strategies. Moreover, hardware-in-the-loop (HiL) simulations are performed to verify the efficiency of the proposed scheme from real-time perspective.