BP神经网络结合粒子群优化卡尔曼滤波的MEMS陀螺随机误差补偿方法  

A Random Error Compensation Method of MEMS Gyroscope Based on BP Neural Network Combined with PSO-Optimized Kalman Filter

作　　者：万芯炜 王晶[1] 杨辉 李毅 张远再 王路 WAN Xinwei;WANG Jing;YANG Hui;LI Yi;ZHANG Yuanzai;WANG Lu(Southwest Institute of Technical Physics,Chengdu 610041,Sichuan,China)

机构地区：[1]西南技术物理研究所,四川成都610041

出　　处：《兵工学报》2023年第2期556-565,共10页Acta Armamentarii

摘　　要：针对微机电系统(MEMS)陀螺仪随机误差相对较大、影响其精度这一问题,提出一种基于BP神经网络结合具有量子行为的粒子群优化(QPSO)算法优化卡尔曼滤波(KF)的补偿方法。采集MEMS陀螺和转台数据作为样本,采用BP神经网络进行训练,建立误差模型;利用训练好的模型对MEMS陀螺进行误差补偿;利用QPSO算法优化KF,以达到更好的降噪效果。实验结果表明,该方法较BP神经网络优化KF、QPSO优化KF与变分模态分解结合小波阈值去噪等方法去噪处理后的平均绝对误差(MAE)和均方误差(MSE)更小,具有更好的降噪效果。To deal with the large random error of the micro-electro-mechanical-system(MEMS) gyroscope that affects its accuracy, an error compensation method based on BP neural network combined with Quantum-behaved Particle Swarm Optimization(QPSO)-optimized Kalman Filter(KF) is proposed. First, the MEMS gyroscope and turntable data are collected as samples, and the BP neural network is employed for training to establish the error model;then the error of the MEMS gyroscope is compensated by the model;finally, the QPSO algorithm is used to optimize KF to achieve better noise reduction effect. The experimental results show that compared with other methods like BP-KF, QPSO-KF and VMD-WTD, this method has better denoising effect, and the MAE and MSE values of the denoised data are smaller.

关 键 词：MEMS陀螺 BP神经网络 量子粒子群优化 卡尔曼滤波 

分 类 号：V241.5[航空宇航科学与技术—飞行器设计]