| With the rapid and healthy development of the social economy and the rapid increase of urban car ownership,the urban building structure is becoming more and more complex,and the positioning based on the shaded environment such as boulevards,viaducts and tunnels becomes more and more complicated.Especially for future areas such as autonomous driving,vehicle collision warning,and visual enhancement of the driving environment,more precise accuracy is required.This paper mainly aims at low-cost,highly reliable vehicle fusion positioning,and combines the complementary advantages of GNSS / INS integrated navigation.The unscented Kalman information fusion algorithm and the time series-based neural network algorithm are proposed to verify that the algorithm can meet GNSS positioning accuracy under long-term failure.It has important theoretical significance and extensive use value for alleviating traffic pressure,improving road traffic safety and reliability,and for fine management and intelligent management of urban traffic.The specific content is as follows:First of all,in order to improve the accuracy and reliability of the urban road vehicle positioning system in my country under the conditions of satellite occlusion.The positionspeed unscented Kalman filter algorithm based on the classic Kalman filter framework is proposed.Based on the GNSS/INS loose combination model,the state equation and measurement equation of the unscented Kalman filter are derived,and the UKF fusion algorithm is given Process.Secondly,in order to further improve the performance of the long-term GNSS signal out-of-lock condition of the urban road intelligent vehicle positioning system,the NARX,LSTM and Elman dynamic neural network models based on time series features are proposed.The INS error will accumulate over time and is susceptible to the vehicle’s motion state,so a high-precision GNSS interrupt time error neural network positioning system structure with external memory feedback function features has been designed,and the NARX,LSTM,Elman network Accurately locate the relationship between input and output.And realized the prediction and compensation of INS error during the long period of GNSS signal interruption.Thirdly,in order to meet the needs of real-time high-precision lane-level intelligent vehicles,satellite loss,and seamless navigation,this paper uses time synchronization and other technologies to develop a set of intelligent,highly refined,and integrated intelligent networked vehicle positioning and navigation systems Including high-precision positioning vehicle system software overall design,hardware design and software function design and technical implementation.The basic platform of hardware equipment built in it can ensure that the vehicle-mounted system can run stably under various complicated road conditions,complete various high-performance satellite positioning and navigation integration,and solve real-time navigation information.The built software platform completed all operations in the entire driving information perception process,which laid the foundation for subsequent experimental performance testing.Finally,through the test results of sports car experiments,it is verified that the unscented Kalman nonlinear filter estimation method and the Elman neural network method based on time series can effectively compensate the position error of the INS under the condition of long-term interruption of the GNSS signal.Among them,the single-step prediction Elman dynamic neural network based on time series achieves the best effect at a step size of 10,and the error proportion of the failure distance in the range of 0.5m and 0.8m reaches 82% in the time period of the signal losing lock for 2min.,100%,the error ratio reaches 51% and 79% in the 5min time period,the maximum error,the average absolute error and the standard deviation reach 81.05 cm,46.25 cm and22.83 cm respectively,achieving slightly better accuracy;based on time series The multistep predictive neural network has a maximum error of 65.92 cm,74.60 cm,and 73.09 cm within 1min,2min,and 5min of the failure time;the average absolute error reaches55.01 cm,74.60 cm,and 73.09 cm,respectively;the standard deviations respectively reach6.19 cm,10.21 cm,6.26 cm,overcoming the position error defects accumulated by INS over time,better robustness and real-time performance than the unscented Kalman algorithm under long-term GNSS failure conditions,compared to the step size of The error rate of the single-step prediction neural network of 10 was increased by 9.8%,3.1%,and 7.13% within 5 minutes of failure,respectively,and the accuracy was further improved.It is verified that the multi-step prediction dynamic neural network based on time series can provide more accurate and reliable continuous positioning information for the case of long-term failure of complex road conditions,and achieve seamless navigation. |
PDF Full Text Request