Stabilized Platform Attitude Control Based on Deep Reinforcement Learning Using Disturbance Observer-Based

Abstract

In order to address the difficulties of attitude control for stabilized platform in rotary steerable drilling, including instability, difficult to control, and severe friction, we proposed a Disturbance Observer-Based Deep Deterministic Policy Gradient (DDPG_DOB) control algorithm. The stabilized platform in rotary steering drilling was taken as a research object. On the basis of building a stabilized platform controlled object model and a LuGre friction model, DDPG algorithm is used to design a deep reinforcement learning controller. After the overall framework of the stabilized platform control system was given, appropriate state vectors were selected, a reward function satisfying the system requirement was designed, an Actor-Critic network structure was constructed and the network parameters was updated. Moreover considering the non-linear friction disturbance that causes steady-state errors, oscillations, and hysteresis phenomena in the stabilized platform control system, a DDPG algorithm based on the disturbance observer was proposed to eliminate the effects of friction disturbance so that to enhance robustness and anti-interference ability of the stabilized platform control system. Experimental results show that the DDPG_DOB control method had good set-point control performance and tracking effect. The tracking error of the tool face angle can be maintained within ± 8.7% and the DDPG_DOB control method can effectively suppress friction interference and improve the nonlinear hysteresis phenomenon when the system is affected by friction interference,enhancing the robustness of the system.