As an emerging air force, UCAV trajectory prediction holds significant importance in preventing and countering adversary threats in the air combat. In order to enhance the performance of UCAV (Unmanned Combat Aerial Vehicles) trajectory prediction as well as the generalization ability of the prediction model, a Dual-Attention GRU method based on deep learning is proposed to predict the UCAV trajectory. We establish a six-degree-of-freedom UCAV model and obtain numerous trajectory samples through simulation. Subsequently, a Dual-Attention GRU model is developed, which employs GRU as the basic unit and dual-stage attention to capture relevant feature information. The input attention stage selects pertinent representations from raw inputs, while the temporal attention stage captures the temporal information of the encoded inputs. The hyperparameter configuration that minimizes the test set loss is identified through experiments. The proposed method is evaluated against FC (Fully Connected Network), GRU (Gate Recurrent Unit Network), LSTM (Long Short-Term Memory Network) and Attention GRU on three evaluation metrics for prediction performance. Experiments indicate that the proposed Dual-Attention GRU trajectory prediction model significantly outperforms other networks in terms of prediction metrics and is capable of meeting the requirements for UCAV trajectory prediction precisely.