Broad Reinforcement Learning for Adaptive Control of a 2-DOF Helicopter System With Unknown Dead Zone
Zhijia Zhao, Yan Weng, Zhijie Liu, Chenguang Yang, C. L. Philip Chen · IEEE Transactions on Industrial Electronics · 2024
In this study, we propose a broad reinforcement learning adaptive control strategy for a two degree-of-freedom helicopter system with unknown dead zone. This strategy utilizes an inverse compensation method to mitigate the impact of the dead zone and incorporates an adaptive parameter system for dead zone error compensation. An action neural network is implemented to approximate model uncertainty, while a critic neural network evaluates the control strategy. The integration of a broad learning system substantially enhances the generalization capabilities and approximation accuracy of the action neural network. Rigorous derivation confirm that all signals in the closed-loop system are semiglobally uniformly ultimately bounded. Both simulation and experimental results validate the effectiveness of the proposed control strategy.