Abstract:In view of the nonlinear and non-stationary characteristics of the ionospheric total electron content(TEC), this paper proposes a short-term ionospheric TEC forecast model, KF-LSTM, based on the long short-term memory(LSTM) neural network. In data processing, Kalman filtering is introduced to preprocess the ionospheric TEC data of the Center for Orbit Determination in Europe(CODE). Meanwhile, the model is used to predict regional ionospheric TEC at 36 grid points in the global high, middle, low latitudes and equatorial regions in 2016 and 2018. The results indicate that the KF-LSTM prediction performance is superior to traditional BP neural network models and LSTM models in different latitude regions. In the equatorial region, its predictive performance is comparable to the C1PG model; in the 15°N-75°N region, the prediction effect is better than the C1PG model.