Welcome to Acta Armamentarii ! Today is Share:

Acta Armamentarii ›› 2022, Vol. 43 ›› Issue (8): 1913-1925.

• Paper •

### TSO-GRU-Ada Maneuver Trajectory Prediction Based on Maneuver Unit Library

TANG Shangqin1, WEI Zhenglei2, XIE Lei1, ZHOU Huan1, ZHANG Zhuoran3

1. （1.Aeronautics Engineering College, Air Force Engineering University, Xi'an 710038, Shaanxi, China;2.China Aerodynamics Research & Development Center, Mianyang 621000, Sichuan, China;3.Unit 93184 of PLA, Beijing 100000, China）
• Online:2022-07-29

Abstract: Maneuver trajectory prediction is an important part of autonomous air combat. To deal with the problem of low accuracy and time-consuming prediction of Unmanned Combat Aerial Vehicle (UCAV) maneuver trajectory, a prediction model based on the gated recurrent neural network with adaptive boosting algorithm integrated with triangle optimization is proposed. Firstly, a three-degree-of-freedom model of the UCAV is established to solve the trajectory data source problem. Secondly, through the four trajectory characteristics, the trajectories are divided into three categories, namely, planar maneuver, spatial left-turning maneuver and spatial right-turning maneuver, and 21 basic maneuver units are constructed. Then the gated recurrent neural network is explained. To prvent the network gradient optimization from falling into the local optimum, the triangle search optimization algorithm is introduced to update the internal weights and biases of the network. At the same time, to improve the prediction accuracy, the adaptive boosting algorithm is used to build a strong predictor. The optimal parameters of the predictor are selected through experiments, and the predictions are made under different maneuver units. The prediction results have high accuracy and all of them can meet the time consumption requirements. Finally, to test the prediction performance for the maneuver trajectory, a trajectory is selected from the Air Combat Maneuvering Instrument, and compared with results from five different prediction models. The results show that the proposed method has the best prediction accuracy.

CLC Number: