基于PSO-WNN的MEMS陀螺随机误差建模普适性分析

Universality Analysis of Stochastic Error Modeling for MEMS Gyroscope Based on PSO-WNN

针对MEMS陀螺仪随机误差对系统导航精度影响以及现有建模方案存在个体普遍适用性问题,提出将粒子群优化算法(PSO)与小波神经网络(WNN)结合后对MEMS陀螺随机误差进行预测的建模方法。完成小波神经网络的构建,利用小波函数作为神经网络中隐含层的激励函数,同时将小波神经网络各层的连接权值作为粒子群优化算法中粒子的位置,使得建立的模型函数逼近能力更加灵活有效且增强其容错能力。同型号不同个体MEMS传感器建模补偿实验结果表明,论文提出的PSO-WNN误差建模方法预测的MEMS陀螺仪随机误差均值和标准差分别优于0.025°/s和0.13°/s;补偿后的MEMS陀螺Allan方差分析结果进一步验证了论文所提方法的可行性与普适性。

According to the system navigation precision was influenced by MEMS gyroscope random errors and the existing modeling scheme were generally applicable. A modeling method for predicting the random error of MEMS gyroscope was proposed by combining particle swarm optimization(PSO)with wavelet neural network(WNN). The wavelet neural network was constructed,and the wavelet function was used as the excitation function of hidden layer in neural network. At the same time,the connection weight of each layer of the wavelet neural network was used as the particle location in the particle swarm optimization algorithm,which makes the model function approximation ability more flexible and effective and enhance its fault tolerance. The experimental results of the modeling compensation of different individual MEMS sensors for the same model showed that,the MEMS gyroscope random error mean and standard deviation were better than 0.025°/s and 0.13°/s based on the PSO-WNN error modeling method. The feasibility and universality of the proposed method was further validated by the compensated MEMS gyroscope Allan variance analysis results.