Research Of Integrated Navigation Algorithm Based On GNSS Pseudo Measurement And Motion State Estimation

In recent years,with the rapid development of smart transportation,the demand for high-precision positioning in complex urban environment is increasingly strong.The INS/GNSS integrated navigation system composed of inertial navigation system and global navigation satellite system can make up for the shortcomings of single navigation system and provide continuous and high-precision navigation information.However,in the actual complex urban environment,GNSS signal is easily blocked by tall buildings,trees,tunnels and etc.Therefore,when the GNSS signal is unavailable,the traditional integrated navigation system degenerates into pure INS.And due to the measurement noise of the inertial measurement units of the micro electronic machine system,the positioning precision of the pure INS diverges rapidly with errors accumulation.To sum up,it is a challenging task to enhance the performance of the INS/GNSS integrated navigation system for land vehicle during GNSS outages.This paper focuses on pseudo GNSS position prediction algorithm and adaptive zero-velocity detection algorithm,and introduces a heterogeneous multi task learning framework with shared "denoising" process.The specific works are as follows:Firstly,in order to bridge GNSS interrupts,an improved pseudo GNSS position estimation algorithm is designed,which leverages the temporal convolutional neural network to directly find out the correlation between INS data and GNSS position increment,provides stable and accurate virtual GNSS position to assist individual INS,and improves the training efficiency.The experimental results show that the improved pseudo GNSS position predictor based on TCN outperforms the existing pseudo position predictors based on LSTM and ALSTM-FCN.Secondly,though existing researches have made reasonable progress in positioning accuracy,they largely ignore sophisticated vehicle stopping events and the further improvement of positioning performance is urgently needed in complex urban environments.This paper develops a robust zero-velocity detection algorithm using one-dimensional deep convolutional neural network to detect the zero-velocity motion sate of vehicles,allowing for timely correcting the velocity and heading.The research further optimizes the problem of unbalanced distribution of positive and negative samples and difficult and easy samples in the classification task.The experimental results show that the proposed zero-velocity detection algorithm based on CNN is superior to the traditional zero-velocity detection algorithm based on variance threshold in the complex road section where the GNSS signal is missing for a long time and the car is stationary for a long time.Finally,considering the correlation between pseudo GNSS position prediction task and zero-velocity detection task,a heterogeneous multi task learning framwork with shared "de-noising" process is introduced to ensure that the whole cascade neural network can jointly learn pseudo GNSS position and zero-velocity update information,so as to further improve the positioning performance of the integrated navigation system when GNSS signals are blocked.The proposed multi-task learning model is evaluated on extensive practical road data,and achieves root mean square position error of 3.794 m for 120 s GNSS outages under long-term vehicle stopping scenarios,which obviously outperforms the stand-alone LSTM,ALSTM-FCN,TCN and TCN+CAE.Experimental results also demonstrate that the proposed MTL method yields a remarkable accuracy over 99.0%for vehicle stationary detection.