基于扩展卡尔曼滤波的多旋翼飞行器融合姿态解算算法

Fusion Attitude Solving Algorithm of Multi Rotor Aircraft Based on Extended Kalman Filter

针对多旋翼飞行器飞行时机身振动对加速度计产生的噪声干扰,建立径向基(RBF)神经网络结构模型,提出扩展卡尔曼滤波融合姿态解算算法。采用加速度计解算的姿态角作为输入向量,四元数互补滤波解算的姿态角作为参考向量,通过学习不断调整网络最优权值,得到滤波后的由加速度计解算的姿态角,并将四元数互补滤波算法解算出的姿态角与经RBF神经网络非线性滤波得到的加速度计解算的姿态角进行扩展卡尔曼滤波融合姿态解算,以提高飞行器姿态角的解算精度。实验结果表明:基于RBF神经网络的非线性滤波算法可对加速度计解算的姿态角进行有效滤波,提高飞行器姿态角的解算精度;采用的扩展卡尔曼滤波融合姿态解算算法收敛速度快、稳定性强,能够更准确地实时解算飞行器的当前姿态,验证了该算法的可行性。

Aiming at the noise disturbance of accelerometer caused by body vibration of multi rotor aircraft during flight,a structure model of radial basis function(RBF)neural network was established,and an extended Kalman filter fusion attitude solving algorithm was proposed.The attitude angles calculated by the accelerometer were used as the input vectors,and the attitude angles calculated by the quaternion complementary filter were used as reference vectors,the optimal weight of network was adjusted continuously by learning,and the attitude angle calculated by accelerometer after filtering was obtained.The attitude angles calculated by the quaternion complementary filter algorithm and the accelerometer solution angles obtained by the nonlinear filtering of the RBF neural network were used to perform the extended Kalman filter fusion attitude calculation and to improve the accuracy of attitude angle calculation.The experimental results show that the nonlinear filtering algorithm based on RBF neural network can effectively filter the attitude angle of the accelerometer solution,and improve the accuracy of the aircraft attitude angle.The proposed extended Kalman filter fusion attitude algorithm has fast convergence and strong stability,and can more accurately calculate the current attitude of the aircraft in real time,and the feasibility of the algorithm is verified.