Trajectory planning optimization and probabilistic inference reinforcement learning vibration control of three translational flexible beams

There are a large number of flexible bodies in spacecraft, which belong to a multi-flexible body system. When the orbit of a spacecraft changes, it will excite persistent and significant vibrations. In order to suppress vibration of the multi-flexible body system, a three translational flexible beam (TTFB) system is constructed and its finite element model is established. A trajectory planning scheme based on bacterial foraging optimization (BFO) algorithm and multi-segment quintic spline interpolation trajectory function is proposed. The optimized three-beam vibration is studied, and the BFO trajectory of three-beam (BFO3B) is obtained. A reinforcement learning (RL) framework is built based on the probabilistic inference model for learning (PIL) and deterministic policy gradient (DPG) algorithm, and the PIL-DPG controller is trained. Experimental results indicate that BFO3B trajectory has better vibration suppression ability than trapezoidal trajectory and cycloidal trajectory. Because of the nonlinearity of the PIL-DPG controller, it has better vibration control ability than proportional–derivative (PD) control.