Abstract:

Ship inertial navigation system (INS) usually integrate with global navigation satellite system (GNSS) to improve its long term position performance. Once GNSS disables, the position error will diverge fast with time. In order to improve INS' long term position accuracy while GNSS disables, a position correction scheme is proposed using back propagation neural network (BPNN) to fit and correct longitude and latitude based on the INS' origin data, and the network's weigh value is updated based on the Bayesian algorithm. While according to the general theory to calculate the number of neurons, the best number of neurons and training sample distribution are determined via a number of experiments. The real ship test results show that while GNSS disables, in the next two hours, neural net correction inertial navigation position obtained by historical data of 24 h is more accurate than the independent-working INS' correction inertial navigation position, the position mean error decrease 63 percent, the maximum error decrease 50 percent and the minimal error decrease to 0.

Key words: inertial navigation system (INS), global navigation satellite system (GNSS) disable, back propagation neural network (BPNN), Bayesian algorithm, position error