Application of Neural Network Aided Particle Filter in GPS Receiver Autonomous Integrity Monitoring

According to the measurement noise feature of GPS receiver and the sample impoverishment problem with the basic particle filter, an improved particle filter based on neural network algorithm is proposed. Using back-propagation (BP) neural network to adjust the particles with too high and too low weight, firstly, the larger weight particles are respectively splitted into two smaller weight particles. Then, abandoning the particles with very small weight, and adjust the particles with smaller weight by using the neural network. Therefore, the diversity of the sample particles is improved. The improved particle filter algorithm is combined with the likelihood ratio method for GPS receiver autonomous integrity monitoring (RAIM). By using the likelihood ratio as a consistency test statistic to achieve the fault detection, satellite fault detection is undertaken by checking the cumulative likelihood ratio of system state with detection threshold. By taking advantage of the relationship in statistical values between the total cumulative likelihood ratio and partial cumulative likelihood ratio, the number of fault satellite can be determined. Based on the real GPS raw data, the simulation results demonstrate that the improved particle filter under the conditions of non-Gaussian measurement noise can effectively detect and isolate fault satellite, and improve the performance of fault detection.