Global ionospheric total electron content prediction system

A global ionospheric total electron content prediction system based on a spatio-temporal sequence hybrid framework, wherein, the system comprises a transmitting station, a space station and a receiver;

• the transmitting station is configured to predict a global ionospheric total electron content and transmit a result of the predicted global ionospheric total electron content to the space station;
• the space station comprises a plurality of satellites for receiving information of the predicted global ionospheric total content and broadcasting the received information of the predicted global ionospheric total content to a global receiver;
• the receiver is configured to receive the information of the predicted global ionospheric total content sent by the space station, and eliminate an error caused by an ionosphere changing a propagation speed and a propagation path of satellite navigation signals according to the information;
• wherein, the transmitting station implements computational processing for two types of spatio-temporal sequences, including a stationary spatio-temporal sequence and a non-stationary spatio-temporal sequence;
• in a case of the stationary spatio-temporal sequence, a STARMA model prediction method is directly used, and the STARMA model prediction method comprises the following steps of:
• (a1) establishing a STARMA model of the stationary spatio-temporal sequence, wherein, firstly, the STARMA model is established and identified:
• an objective function of the STARMA model is firstly established, and then autocorrelation and partial correlation functions of the STARMA model are used to perform model identification and determine a function order;
• secondly, model parameters of the STARMA model are estimated;
• (a2) checking a residual of the STARMA model after the STARMA model is established;
• and
• (a3) estimating a precision of the predicted result of the model by using a root mean square error (RMS) and an average error (Bias);
• in a case of the non-stationary spatio-temporal sequence, a ConvLSTM+STARMA hybrid model prediction method is used, which comprises the following steps of:
• (b1) establishing a hybrid model comprising a model portion A and a model portion B, wherein firstly, the model portion A is to extract a spatial nonlinear trend of the non-stationary spatio-temporal sequences by using Convolutional Neural Networks, extract a temporal nonlinear trend of the non-stationary spatio-temporal sequence by using a LSTM neural network, and obtain a non-linear trend model of the non-stationary spatio-temporal sequence by the two extraction;
• secondly, a nonlinear trend in data is obtained using the model portion A, the nonlinear trend in the data is removed to obtain residual data, and a stationarity test is carried out on the obtained residual data;
• if the test fails, the model portion A is readjusted to extract the nonlinear trend again until the residual data pass the stationarity test;
• and thirdly, the model portion B is to perform STARMA modeling on stationary residual data if the test is passed;
• (b2) performing a residual test on a fitting result of the hybrid model;
• and
• (b3) after passing the residual test, predicting the global ionospheric total electron content using the hybrid mode, and evaluating the prediction accuracy of the predicted result using the root mean square error (RMS) and average error (Bias).