Random Error Source Identification of SINS by LSTM Hubrid Convolution Network

For improving traditional methods generalization ability of identifying random error sources in aircraft strap-down inertial navigation(SINS) and applied online in real time, through the design of deep learning network to replace the concrete modeling method combined practical problem, the output signal characteristics are extracted by convolution, design the softmax layer to identify the error source significance, and actually build the sample set and test set to verify the network. The results show that the deep learning network can meet the requirements of online analysis, although the results are somewhat crude compared to the traditional methods, but it can reflect the system state in real time, which can provide the basis for the optimization of navigation filtering algorithm.