When satellite signals pass through the ionosphere, the total electron content (TEC) is the main physical quantity that causes errors in the ionosphere, Moreover, the short-term prediction of TEC is an important research object in the field of ionospheric monitoring, and the accurate calculation and short-term prediction of ionospheric TEC is one of the most important elements of space weather monitoring and early warning. In this paper, a new crow search algorithm (CSA) is proposed to improve the LSTM neural network by optimizing the initial weights and parameters of long and short-term memory neural network, and to forecast the TEC values in the future period. The experimental results show that in the prediction of the ionosphere at single station for 48 h in 2023 in the Chinese region, the forecast effect at mid-latitudes; the CSA-LSTM forecast model is 0.72 TECu smaller than the average root-mean-square error of the single LSTM forecast model, while the forecast effect at low latitudes; the CSA-LSTM forecast model is smaller than the average root-mean-square error of the single LSTM forecast model by 2.67 TECu, overall, the CSA-LSTM forecast model has higher forecast accuracy and better forecasts at lower latitudes.