基于LSTM-MPC的四旋翼无人机不平衡负载姿态控制
方应才1，张东升1,2*
1.山东交通学院工程机械学院，山东 济南 250357；
2.山东省交通建设装备与智能控制工程实验室，山东 济南 250357
摘要：为解决四旋翼无人机姿态控制中不平衡负载及控制系统非线性问题，采用长短期记忆（long short-term memory, LSTM）神经网络与模型预测控制（model predictive control, MPC）优势互补方法，提出LSTM-MPC策略。通过LSTM神经网络预测姿态变化，增强无人机姿态控制系统对误差的预判能力；将MPC作为前馈控制，动态优化控制输入，二者结合显著提高系统控制精度。采用MATLAB对四旋翼无人机不平衡负载姿态控制进行仿真试验，仿真结果表明：采用LSTM-MPC策略对四旋翼无人机滚转角、俯仰角和偏航角期望值跟踪效果的均方根误差比采用MPC策略分别减小13.33%、12.31%和11.11%，比采用模糊比例-积分-微分（proportional-integral-derivative, PID）策略分别减小14.05%、25.33%、23.81%。采用某品牌F450四旋翼无人机平台搭载0.6 kg负载开展不平衡负载姿态控制飞行测试，测试结果表明：采用LSTM-MPC策略的四旋翼无人机滚转角、俯仰角和偏航角实际输出结果与期望值的平均误差分别为3.91%、5.31%和1.10%，表明LSTM-MPC策略能有效提高四旋翼无人机不平衡负载姿态控制的飞行稳定性。
关键词：四旋翼无人机；LSTM神经网络；MPC；不平衡负载；姿态控制
Attitude control of quadrotor UAV with unbalanced load based on LSTM-MPC
FANG Yingcai1, ZHANG Dongsheng1,2*
1. School of Construction Machinery, Shandong Jiaotong University, Jinan 250357, China;
2. Shandong Provincial Engineering Laboratory for Transportation Construction Equipment and Intelligent Control, Jinan 250357, China
Abstract: To address the issues of unbalanced load and system nonlinearity in quadrotor unmanned aerial vehicles（UAV）attitude control, a LSTM-MPC strategy is proposed by combining the advantages of long short-term memory (LSTM) neural network and model predictive control (MPC). LSTM neural network is used to predict attitude changes, enhancing the system′s ability to anticipate errors. MPC is employed as feedforward control to dynamically optimize control inputs. The combination significantly improves system control accuracy. MATLAB simulation experiment on quadrotor UAV attitude control with unbalanced load shows that: compared to MPC strategy, the LSTM-MPC strategy reduces the root mean square error of tracking expected values for roll angle, pitch angle, and yaw angle by 13.33%, 12.31%, and 11.11% respectively; compared to fuzzy PID strategy, it reduces by 14.05%, 25.33%, and 23.81% respectively. Flight test is conducted using a branded F450 quadrotor UAV platform carrying a 0.6 kg load to test unbalanced load attitude control. The test result shows that the average errors between the actual output and expected values of the quadrotor UAV′s roll, pitch, and yaw angles using the LSTM-MPC strategy are 3.91%, 5.31%, and 1.10%, respectively, indicating that the LSTM-MPC strategy can effectively improve the flight stability of quadrotor UAV attitude control with unbalanced load.
Keywords: quadrotor UAV; LSTM neural network; MPC; unbalanced load; attitude control
