基于GRU循环神经网络辅助的组合导航算法

Integrated navigation algorithm based on GRU cyclic neural network

针对复杂场景下GPS信号失锁导致的INS/GPS组合导航系统定位精度严重下降问题,提出基于GRU(门控循环单元)循环神经网络辅助的方法。在GPS信号锁定的情况下,使用GRU循环神经网络对IMU传感器数据、组合导航信息、GPS信息进行训练;GPS信号失锁后,利用训练好的模型进行预测,继续补偿INS结果。通过实际跑车采得的数据进行验证,在GPS信号失锁时,使用GRU循环神经网络辅助相较于纯惯导系统精度有较大提高。与MLP(多层感知器)辅助的方法进行比较,验证了循环神经网络对于连续时间轨迹推算的优越性。

Aiming at the problem that the positioning accuracy of INS/GPS integrated navigation system decreases seriously due to GPS signal unlocking in complex scenes,a method based on GRU recurrent neural network(RNN)was proposed.When GPS signal was locked,GRU recurrent neural network was used to train IMU sensor data,integrated navigation information and GPS information.During the GPS outages,the trained model was used to predict and continue to compensate for the result of INS.Experimental results show that during the GPS outages,the accuracy of GRU recurrent neural network is greatly improved compared with that of the pure inertial navigation system.Moreover,compared with MLP(multilayer perceptron)assisted method,the superiority of recurrent neural network for continuous time trajectory estimation is verified.