Aiming at the problem that civil aviation GPS clock error is susceptible to interference and deception signals, an improved CNN-BiLSTM-Attention prediction model is proposed. The dynamic median absolute deviation method is used to eliminate the single-difference sequence outliers of the clock error, and the cubic spline interpolation method is used to repair the data. Based on the key features extracted from the original data, the anti-interference ability of the CNN-BiLSTM-Attention prediction model with and without interference, as well as short-term (1 h), medium-term (6 h) and long-term (12 h) were evaluated by ablation experiments. The results show that compared with ARIMA, the residual prediction deviation of CNN-BiLSTM-Attention prediction model is increased by 49.2%; compared with the XGBoost model, the mean absolute error is reduced by 53.3%; compared with the ELM model, root mean square error decreased by 51.2% and R2 increased by 34.7%. The model has high prediction accuracy and stability in complex environments, and can significantly improve the prediction performance of GPS clock error, which provides a new approach for high-precision GPS anti-jamming positioning.
动态中位数绝对偏差(dynamic median absolute deviation,DMAD)通过计算单差序列中每个数据点的绝对偏差值,并根据中位数计算,基于每个数据点与中位数的偏差是否超过设定的阈值来识别异常值并标记。在设置阈值系数时,原则上保证检测出的异常值不超过数据总量的5%,避免对有效数据的误删。一旦检测到异常值,使用三次样条插值法替换异常值,利用相邻的非异常值生成平滑、连续的序列,减少异常值对模型预测的影响。
设原始时钟误差数据为,其中,Xi 表示第i个历元的时钟误差。对 X 进行一阶差分,生成单差序列,以突出局部异常点。单差序列中包含随趋势项变化数据,这些趋势项会对离群值检测产生干扰,一阶差分公式为
LORRAINEK J S, RAMARAKULAM. A comprehensive survey on GNSS interferences and the application of neural networks for anti-jamming[J]. IETE Journal of Research, 2023, 69(7): 4286-4305.
BHATTA, MEHTAI, PATIDARP. Leveraging LSTM for predictive modeling of satellite clock bias[C]//2025 8th International Conference on Data Science and Machine Learning Applications. February 16-17, 2025, Riyadh, Saudi Arabia. IEEE, 2025: 25-30.
[4]
VAN HOUDTG, MOSQUERAC, NÁPOLESG. A review on the long short-term memory model[J]. Artificial Intelligence Review, 2020, 53(8): 5929-5955.
[5]
ZHANGF, NIUY T, ZHOUQ, et al. Intelligent anti-jamming decision algorithm for wireless communication under limited channel state information conditions[J]. Scientific Reports, 2025, 15: 6271.
[6]
XIEY C, CHENL, ZHANGK, et al. Deep neural network based anti-jamming processing in the GNSS array receiver[J]. GPS Solutions, 2025, 29(2): 63.
[7]
LIB Y, QIAOJ, LUZ K, et al. Influence of sweep interference on satellite navigation time-domain anti-jamming[J]. Frontiers in Physics, 2023, 10: 1063474.
[8]
WANGY P, LUZ P, QUY Y, et al. Improving prediction performance of GPS satellite clock bias based on wavelet neural network[J]. GPS Solutions, 2017, 21(2): 523-534.
[9]
ZHANGG C, HANS H, YEJ, et al. A method for precisely predicting satellite clock bias based on robust fitting of ARMA models[J]. GPS Solutions, 2021, 26(1): 3.
[10]
HES F, LIUJ L, ZHUX W, et al. Research on modeling and predicting of BDS-3 satellite clock bias using the LSTM neural network model[J]. GPS Solutions, 2023, 27(3): 108.
[11]
CAIC L, LIUM Y, LIP C, et al. Enhancing satellite clock bias prediction in BDS with LSTM-attention model[J]. GPS Solutions, 2024, 28(2): 92.
[12]
JINL H, ZHAOW Z, PANX, et al. Combining semiparametric and machine learning approaches for short-term prediction of satellite clock bias[J]. Scientific Reports, 2025, 15: 11880.
[13]
SONGL J, LIUJ H, WANGL L, et al. A comparative study on different machine learning approaches with periodic items for the forecasting of GPS satellites clock bias[J]. Scientific Reports, 2025, 15: 2709.
[14]
HUANGB H, JIZ X, ZHAIR J, et al. Clock bias prediction algorithm for navigation satellites based on a supervised learning long short-term memory neural network[J]. GPS Solutions, 2021, 25(2): 80.
WUChunjun, SUNYueqiang, WANGXianyi, et al. Assessment of position performance of BDS for space application based on FY-3D satellite[J]. Geomatics and Information Science of Wuhan University, 2023, 48(2): 248-259.
[17]
ZHONGM, HUM N, HUF, et al. An improved lightweight YOLO algorithm for recognition of GPS interference signals in civil aviation[J]. IET Signal Processing, 2024, 2024(1): 9927636.
RENJianji, WEIHuihui, ZOUZhuolin, et al. Ultra-short-term power load forecasting based on CNN-BiLSTM-Attention [J]. Power System Protection and Control, 2022, 50(8): 108-116.
[20]
LUOS C, WANGB S, GAOQ Z, et al. Stacking integration algorithm based on CNN-BiLSTM-Attention with XGBoost for short-term electricity load forecasting[J]. Energy Reports, 2024, 12: 2676-2689.
[21]
WANGX Y, YANGJ J, HUANGM, et al. GNSS interference and spoofing dataset[J]. Data in Brief, 2024, 54: 110302.
[22]
GEH B, LIB F, WUT H, et al. Prediction models of GNSS satellite clock errors: evaluation and application in PPP[J]. Advances in Space Research, 2021, 68(6): 2470-2487.
[23]
YAS S, ZHAOX W, LIUC, et al. Improved BDS-2/3 satellite ultra-fast clock bias prediction based with the SSA-ELM model [J]. Sensors, 2023, 23(5): 2453.
[24]
SYAMW P, PRIYADARSHIS, ROQUÉA A G, et al. Fast and reliable forecasting for satellite clock bias correction with transformer deep learning[C]//Proceedings of the 54th Annual Precise Time and Time Interval Systems and Applications Meeting,January 23-26, 2023, Long Beach, California. Institute of Navigation, 2023: 76-96.
[25]
LID A, MAX L, ZHAOJ M,et al. Mitigating GNSS multipath effects using XGBoost integrated classifier based on consistency checks[J]. International Journal of Antennas and Propagation, 2022, 2022: 2742620.