Model Reduction and Chattering Mitigation in Multi-Model Predictive Control for Quadrotor UAVs

This study introduces a multi-model predictive control (MMPC) framework for quadrotor attitude regulation, combining the high-precision capabilities of nonlinear model predictive control (NMPC) with the computing efficiency of linear model predictive control (LMPC). Although NMPC provides excellent regulation, its substantial computational expense restricts its real-time use in high-refresh-rate applications like attitude control. In contrast, LMPC exhibits computational efficiency but encounters challenges with stability and precision in nonlinear dynamics. In order to resolve this discrepancy, we propose a Multi-Model Predictive Control (MMPC) approach that integrates Principal Component Analysis (PCA) for model reduction and employs linear quadrotor attitude models. This approach reduces the computational complexity and the number of necessary Linear Model Predictive Controllers (LMPCs). To improve performance, we implement an adaptive gain scheduling algorithm to mitigate the chattering effect frequently seen in multi-model approaches. This seamless switching technique diminishes high-frequency noise in brushless DC motor operation, minimises peak overshoot, and mitigates actuator wear, thereby enhancing system stability and durability. Comparative assessments indicate that the suggested MMPC framework attains NMPC-equivalent performance while incurring LMPC-level computational expenses, exceeding the efficacy of conventional incremental nonlinear dynamic inversion, sliding mode control, and traditional NMPC and LMPC methodologies. This study offers a computationally efficient and stable control solution for real-time UAV applications, including model reduction, chattering mitigation, and predictive control optimisation for autonomous aerial systems.