关键词: 多旋翼飞行器, LSTM神经网络, 自适应滑模控制器, PSoC架构

Abstract: This paper addresses the issues of chattering and response lag in multi-rotor aircraft under complex wind field conditions by proposing a neural network-enhanced adaptive sliding mode control method based on a PSoC hardware platform. The method replaces the fixed disturbance upper-bound estimation model in traditional adaptive sliding mode control with a long short-term memory network, enabling dynamic optimization of the sliding mode parameters λ and ρ. Furthermore, an efficient deployment of the algorithm is achieved using the heterogeneous computing architecture of PSoC with an integrated NPU. Hardware acceleration performance tests demonstrate that the PSoC hardware platform requires only 81.7 μs for single-sample processing, supporting high-frequency closed-loop control at 12.24 kHz. Simulation results indicate that, compared to ASMC, the proposed method reduces the convergence time of the system´s pitch/roll angles from 0.62 s to 0.27 s, suppresses the roll angle buffeting amplitude from ±1.66° to ±0.14°, and maximum movement speed within is ±1.1 m/s.

Key words: multi-rotor aircraft, LSTM neural networks, ASMC, PSoC architecture