Wu You, Yu Hongyi, Du Jianping, Ge Chenglong
Information System Engineering College, PLA Strategic Support Force Information Engineering University, Zhengzhou 450001, China.
Sensors (Basel). 2023 Nov 11;23(22):9120. doi: 10.3390/s23229120.
In the realm of aviation, trajectory data play a crucial role in determining the target's flight intentions and guaranteeing flight safety. However, the data collection process can be hindered by noise or signal interruptions, thus diminishing the precision of the data. This paper uses the bidirectional encoder representations from transformers (BERT) model to solve the problem by masking the high-precision automatic dependent survey broadcast (ADS-B) trajectory data and estimating the mask position value based on the front and rear trajectory points during BERT model training. Through this process, the model acquires knowledge of intricate motion patterns within the trajectory data and acquires the BERT pre-training Model. Afterwards, a refined particle filter algorithm is utilized to generate alternative trajectory sets for observation trajectory data that is prone to noise. Ultimately, the BERT trajectory pre-training model is supplied with the alternative trajectory set, and the optimal trajectory is determined by computing the maximum posterior probability. The results of the experiment show that the model has good performance and is stronger than traditional algorithms.
在航空领域，轨迹数据在确定目标飞行意图和保障飞行安全方面发挥着至关重要的作用。然而，数据收集过程可能会受到噪声或信号中断的阻碍，从而降低数据的精度。本文使用来自变换器的双向编码器表示（BERT）模型来解决该问题，即在BERT模型训练期间，通过对高精度自动相关监视广播（ADS-B）轨迹数据进行掩码处理，并根据前后轨迹点估计掩码位置值。通过这个过程，模型获取轨迹数据中复杂运动模式的知识并获得BERT预训练模型。之后，利用改进的粒子滤波算法为容易受到噪声影响的观测轨迹数据生成替代轨迹集。最终，将替代轨迹集提供给BERT轨迹预训练模型，并通过计算最大后验概率来确定最优轨迹。实验结果表明，该模型具有良好的性能，并且比传统算法更强。
