INS-GNSS Integrated Navigation Algorithm Based on TransGAN

With the rapid development of autopilot technology, a variety of engi-neering applications require higher and higher requirements for navigation and positioning accuracy, as well as the error range should reach centimeter level. Single navigation systems such as the inertial navigation system (INS) and the global navigation satellite system (GNSS) cannot meet the navigation require-ments in many cases of high mobility and complex environments. For the purpose of improving the accuracy of INS-GNSS integrated navigation system, an INS-GNSS integrated navigation algorithm based on TransGAN is proposed. First of all, the GNSS data in the actual test process is applied to establish the data set. Secondly, the generator and discriminator are constructed. Borrowing the model structure of generator transformer, the generator is constructed by multi -layer transformer encoder, which can obtain a wider data perception ability. The generator and discriminator are trained and optimized by the production countermeasure network, so as to realize the speed and position error compensa-tion of INS. Consequently, when GNSS works normally, TransGAN is trained into a high-precision prediction model using INS-GNSS data. The trained Trans-GAN model is emoloyed to compensate the speed and position errors for INS. Through the test analysis of flight test data, the test results are compared with the performance of traditional multi-layer perceptron (MLP) and fuzzy wavelet neural network (WNN), demonstrating that TransGAN can effectively correct the speed and position information when GNSS is interrupted, with the high accuracy.