Research On GPS/DR Position Method For Vehicle State In Driving Conditions

With the increasing complexity of urban transportation systems and the increasing number of vehicles, vehicle positioning system applications are growing in popularity. The widely used Global Positioning System (GPS) has the advantages of positioning speed with high precision and continuous ground coverage, but occluders such buildings will affect the reception of the GPS signal can cause failure in positioning; dead reckoning (DR) uses vehicle sensors information to estimate vehicle instantaneous location, which can work independently and has high positioning accuracy in short term, but it is not satisfactory in long term because the positioning error will diverge over time. Combining GPS and DR can make them the two complement each other and achieve better positioning performance. The core of the combination is fusion algorithm of multi-sensor data. Kalman filter algorithm uses the principle of information distribution to achieve the optimal multi-sensor information integration and improve the positioning accuracy of GPS/DR system.However, the existing vehicle position and navigation technologies just focus on the spatial location of moving vehicles, while ignore the role of its time information, that is, ignore the impact to location state by driving conditions, traffic conditions and time delay of the positioning system, which will lead to large positioning error and even wrong orientation and serious traffic accidents. To this end, Aiming at urban driving conditions, research on GPS/DR position principle and method for vehicle state in driving conditions is carried out.Firstly, according to the theory of standard linear Kalman filter, extended Kalman filter mathematical models of GPS and DR are deduced separately. Without reset mode two-step federal Kalman filter is designed. According to the size of data error of GPS and DR, a data fusion scheme with adaptive fusion coefficients is designed to integrate the positioning data of GPS/DR system and obtain precise location of the driving vehicle. With the fused location data, the concept of driving conditions is innovatively introduced to the research of vehicle GPS/DR positioning. The road spectrum information collected is learned by the learning function of neural network. With the fused location data as input, the neural network forecasts the positioning data of next moment. In this way, positioning error causing by time delay is corrected, and the accuracy and real-time ability are improved. Single point positioning experiments in 3 different representative working conditions and vehicle dynamic positioning experiments in driving cycles are carried out with a GPS module. Experiment results indicate the instability of positioning accuracy in urban driving conditions with GPS working alone. Meanwhile, simulations of GPS/DR neural network prediction based on Matlab are performed and discussions on positioning errors and its improvement are performed.Finally, the dissertation is summarized and some precious proposal is put forward to the next research.