237 / 2023-09-29 17:20:20

Universality Analysis of MEMS Gyro Random Error Modeling Based on PSO-WNN

Wavelet Neural Network,MEMS,excitation function

The limitations of the device processing result in random errors being the main factor limiting the output accuracy of MEMS gyroscopes. Scholars at home and abroad have long been conducting research on highly feasible and reliable error compensation methods. Due to the rapid development of neural networks, which makes its corresponding research value in gyro drift prediction. Some researchers have also been trying to invoke particle filtering, wavelet neural networks (WNN), and other methods into the error compensation of MEMS gyroscopes.Although wavelet neural networks have many advantages of both wavelet analysis and neural networks, there are some problems in practical applications. For example, the wavelet basis function, the number of nodes in the hidden layer of the neural network can only be selected empirically and cannot be adaptively selected. This tends to cause the network to converge slower, take longer to train, and fall into local minima. Aiming at the impact of MEMS gyroscopes random errors on the system navigation accuracy and the general applicability of the existing modeling scheme, a modeling method is proposed to predict the MEMS gyroscopes random errors after combining the Particle Swarm Optimization (PSO) algorithm with the Wavelet Neural Network (WNN).

 To complete the construction of a wavelet neural network, using the wavelet function as the excitation function of the hidden layer in the neural network, and at the same time, the connection weights of each layer of the wavelet neural network are used as the positions of the particles in the particle swarm optimization algorithm, which makes the function approximation ability of the established model more flexible and effective, and strengthens its fault-tolerance ability.

The wavelet neural network method is used to model the random error of MEMS gyroscope, which mainly includes three parts: network construction, training and testing. Although the MEMS gyroscope random error has a great randomness, the three-layer network structure has a good fitting effect on the nonlinear function and the network structure is simple .

In order to verify the feasibility of the proposed scheme in this paper, the same type of MEMS inertial measurement unit existing in the laboratory is placed behind the antistatic experimental bench, and the output information of the MEMS gyroscope is collected after being powered on for 1 hour, and the 100000 coaxial output signal data of the MEMS gyroscope are intercepted as the test samples, and the wavelet neural network is utilized under the optimization of the PSO algorithm to carry out the modeling and error prediction.