Research On Application Of Particle Filter Algorithm Based On Neural Network In GPS

Since USA built the Global Positioning System (GPS) in1994,GPS currently has beenapplied more and more widely in military and civilian fields, technology trends in componentminiaturization and large-scale manufacturing also led to GPS receiver deeply embedded intomany of the items in our daily life, for example, the mobile phone and the car, which aregrowing popularity now. It would subject to various errors which affects the accuracy of GPSpositioning in the GPS positioning solution, and the common method to reduce errors is thedynamic filtering, but positioning state model of the GPS receiver is generally nonlinearsystem with non-Gaussian noise, and thus need a filtering method which suits for nonlinearsystem with non-Gaussian noise. Because of the advantages of the particle filter in processingnonlinear system with non-Gaussian noise issues, this paper introduces particle filter into theprocessing of GPS data.Aiming at the particles degeneracy in particle filter and samples diversity loss asresampling, using the excellent nonlinear mapping ability of neural network, this paperintroduced BP Neural Network and General Regression Neural Network to improve particlefilter. This paper researched the particle filter of importance weight adjustment based on BPneural network (NNWA-PF), adjust the weights of particles, in order to improve the samplesdiversity and reduce particle degradation. Then this paper introduced General regressionneural network, researched the particle filter of importance state adjustment based on Generalregression neural network (NNISA-PF) adjust the state of particle, in a word, the importancedensity function of particle filter is optimized. Simulation results based on MATLAB provedthat both two particle filters based on BP neural network and GRNN are available andeffective.On the other hand, the positioning precision of GPS at present has basically met theactual requirements of most users, but stringent requirements for its reliability and integrityare demanded. Therefore, this paper utilized these two particle filter algorithm based on BP neural network and GRNN to improve the GPS Receiver Autonomous Integrity Monitoring(RAIM), established the GPS satellite Fault Detection and Isolation (FDI) model, two faulttypes are established named bias error and ramp error, detected and isolated the faulty satellitethrough applying the log-likelihood ratio method, gathered the measured data by the GPSreceiver. Then this paper compared the simulation results based on MATLAB with the resultsof FDI method which uses the basic particle filter. Simulation results show that the proposedFDI algorithm based on NNWA-PF and NNISA-PF successfully detected and isolated thefaulty satellite in the case of non-Gaussian measurement noise, and the introduction of neuralnetwork improves the accuracy of fault detection and isolation both for bias error and ramperror, proves the usability and effectiveness of applying neural network into particle filteralgorithm based on neural network in the GPS RAIM.