In this article, a new long short-term memory (LSTM) network with horizontal heading change modeling is proposed to predict heading and position of large civil aircraft. In data segmentation and preprocessing, a smoothing filter method with forward sliding processing is used, and the smoothing filter ensures continuity of the trajectory. Our heading change modeling encodes aircraft heading information in relation to the nearest waypoints. The encoded vectors are used as the input part of the prediction network, which improves the accuracy of heading prediction. The higher the accuracy of heading prediction, the better the results of cross-distance prediction, and therefore the better the results of position prediction. For practical comparison, we used the route of C919 and related flight data for trajectory prediction experiments. The quantitative analysis results comparing with LSTM network, gate recurrent unit network and embedded heading change modeling show that our model outperforms the state-of-the-art models mentioned above, which demonstrates that heading change modeling can indeed improve the trajectory prediction accuracy. Meanwhile, in the experimental results session, this paper also analyses how the improved accuracy of heading prediction affects the position prediction, further supporting the effectiveness of heading change modeling.