Genetic Wavelet Neural Network And Its Application In Navigation Sensor

Under the development of aviation and navigation technologies, precise navigation system is becoming to indispensable equipment in the carrier. However, no navigation sensor can be fit for all situations. As the result, the individual navigation sensors, which provide position parameters and other information, had developed to the combination of navigation system and measurement system. It is a multi-sensor integrated navigation system which can provide accurate position and military measurement. In that case, fault diagnosis technic is becoming more important, and intelligent diagnosis techinic emerging infuses activity into traditional diagnosing technology.Based on navigation sensors, an intelligent diagnosing network combined wavelet transformation, genetic algorithm and neural network has been proposed in. this paper. After introducing several kinds of combination modes of integrated navigation systems, the operation principle of typical sensors such as gyro, GPS, DVL has been discussed with their error factors. In addition, the fault patterns of gyro sensor have been detailed.After comparing wavelet transform and Fourier transformation based on the characters of gathered signal, the theory of wavelet packet has been researched with discussion of wavelet transform properties in time and frequency domains. By analyzing particular signal, wavelet packet indicated high resolution in high frequency and free further decomposing ability in choosing frequency band, which are advantageous to analyze fault sensor signal. Simulation aiming at gyro signal which is the key of inertia system has been put forward. Sampling and simulating outputs of gyro in different fault models, the signal was then decomposed by three-layer wavelet packet. 8-dimensional eigenvectors have been calculated as the frequency distribution mapping diversified fault modes. Then, network was trained with the eigenvectors samples with high accuracy.After fault eigenvectors had obtained, diagnosing network was designed. Based on the research of Neural Network in fault diagnosis, we introduced Genetic Algorithm to overcome the constringency deficiency of widely-used Back Propagation network. Ergodic initial value of weight and bias is searched for further training by introducing Advanced Genetic Algorithm, which improves the search efficiency and global convergence of the network. By comparing the simulation of BP, GA-BP and Advanced GA-BP, results show that the convergence velocity and training rate have been improved by 50%. Eigenvectors samples decomposed by wavelet packet were used to train the Advanced Genetic Wavelet Network based on sytem framework. Network testing and diagnosis results show the improved network has high accuracy up to 98%.