Satellite Telemetry Parameter Prediction Based on Improved Combinatorial Machine Learning
-
摘要: 对在轨卫星的运行状态进行监测、分析以及异常检测是卫星在轨运行管理的重要内容。预测卫星遥测参数序列的变化趋势，对卫星异常检测与处置、保障安全运行非常必要。针对目前对于周期性不明显且具有多种变化特征的遥测参数预测精确度不够的问题，本文引入对遥测参数的预测有辅助作用的因素作为协变量，提出了基于改进组合机器学习的预测模型。该模型使用全局模型和局部模型分别获取遥测参数序列的趋势特征和局部不规则波动特征，并采用改进的注意力机制捕获多维参数之间的关联关系，提高了预测精度。此模型可以提供点预测和区间预测的结果，为在轨卫星处置决策提供了更多输入。在科学卫星真实遥测数据集和时间序列公开数据集上验证了本文方法的有效性。Abstract: The monitoring, analysis and anomaly detection of the operational status of satellites in orbit are important contents of satellite operational management. It is very necessary to predict the changing trend of satellite telemetry parameter data series for detecting, dealing with satellite anomalies and ensuring the safe operation of satellites. Aiming at the problem that the current prediction research is not accurate enough for telemetry parameters with insignificant periodicity and multiple changing characteristics, this paper introduces covariates that are helpful for telemetry parameter sequence prediction, and proposes a prediction model based on improved combined machine learning, using the global model and the local model to obtain the trend characteristics and local irregular fluctuation characteristics of the telemetry parameter sequence respectively, and the improved Attention mechanism is used to capture the correlation between multi-dimensional parameters to improve the prediction accuracy. At the same time, this model can provide point prediction and interval prediction results for the telemetry data sequence, providing more input for the decision-making of on-orbit satellite disposal. The validity of the proposed method is verified on the real telemetry data set of scientific satellites and the public time series data sets.
-
图 1 卫星遥测参数序列预测模型框架
Figure 1. Framework of satellite telemetry parameter sequence prediction model
图 2 预测30个时刻（a）和50个时刻（b）的结果（P10 quantile和P90 quantile分别代表区间预测的下界和上界，P50 forecast代表点预测结果，true代表真实数据）
Figure 2. Result graph of 30 predictions (a) and 50 predictions (b) (P10 quantile and P90 quantile represent the lower and upper bounds of the prediction interval, P50 forecast represents point prediction results and true is the true value)
表 1 数据集统计表
Table 1. Statistics of all datasets
数据集 维度 样本量 采样率/min 数据类型 Satellite 4 55156 0.5 float Solar energy 2 52560 5 float Electricity 2 105215 15 float 
下载: 导出CSV
表 2 各数据集对应算法的MAPE
Table 2. Each dataset corresponds to the MAPE of algorithms
Model Solar energy Electricity Satellite ARIMA 0.141394 0.364261 0.43033 SVR 0.245441 0.458509 0.46821 GPR 0.126012 0.468745 0.44977 LSTM 0.121016 0.252913 0.32426 Our Model 0.110150 0.145592 0.17361
下载: 导出CSV
表 3 各数据集对应算法的PID
Table 3. Each dataset corresponds to the PID of algorithms
Model Solar energy Electricity Satellite VAR 0.36534 0.34288 0.45087 RVM 0.33673 0.32538 0.39721 GPR 0.42383 0.40293 0.42032 本文模型 0.31160 0.23165 0.20249
下载: 导出CSV
表 4 消融实验结果
Table 4. Results of ablation experiment
MAPE PID 去掉改进Attention结构 0.28864 0.34320 去掉全局–局部融合结构 0.30388 0.40542 LSTM 0.32426 0.42310 本文模型 0.19161 0.23744
下载: 导出CSV
-
[1] 谭春林, 胡太彬, 王大鹏, 等. 国外航天器在轨故障统计与分析[J]. 航天器工程, 2011, 20(4): 130-136TAN Chunlin, HU Taibin, WANG Dapeng, et al. Analysis on foreign spacecraft in-orbit failures[J]. Spacecraft Engineering, 2011, 20(4): 130-136 [2] 庞景月. 基于概率性预测的航天器遥测数据异常检测方法研究[D]. 哈尔滨: 哈尔滨工业大学, 2019PANG Jingyue. Probabilistic prediction based anomaly detection method for spacecraft telemetry data[D]. Harbin: Harbin Institute of Technology, 2019 [3] WILSON G T. Time series analysis: forecasting and control, 5 th Edition, by George E. P. Box, Gwilym M. Jenkins, Gregory C. Reinsel and Greta M. Ljung, 2015. Published by John Wiley and Sons Inc., Hoboken, New Jersey, pp. 712. ISBN: 978-1-118-67502-1[J]. Journal of Time Series Analysis, 2016, 37(5): 709-711 doi: 10.1111/jtsa.12194 [4] ÖLLER L E, STOCKHAMMAR P. Rob J. Hyndman, Anne B. Koehler, J. Keith Ord, Ralph Snyder, forecasting with exponential smoothing: the state space approach, Springer (2008), 359 pp, ISBN 978-3-540-71916-0 (paperback), €36.95, e-ISBN 978-3-540-71918-2 (online), €25/Chapter[J]. International Journal of Forecasting, 2010, 26(1): 204-205 doi: 10.1016/j.ijforecast.2009.09.005 [5] ROMAN D, SAXENA S, ROBU V, et al. Machine learning pipeline for battery state-of-health estimation[J]. Nature Machine Intelligence, 2021, 3(5): 447-456 doi: 10.1038/s42256-021-00312-3 [6] O'MEARA C, SCHLAG L, WICKLER M. Applications of deep learning neural networks to satellite telemetry monitoring[C]//Proceedings of the 2018 SpaceOps Conference. Marseille, France: AIAA, 2018 [7] 李志强, 张香燕, 田华东. 应用HP滤波的卫星遥测数据预测方法[J]. 航天器工程, 2021, 30(4): 23-30LI Zhiqiang, ZHANG Xiangyan, TIAN Huadong. Prediction method of satellite telemetry data using HP filter[J]. Spacecraft Engineering, 2021, 30(4): 23-30 [8] LIU D T, PANG J Y, SONG G, et al. Fragment anomaly detection with prediction and statistical analysis for satellite telemetry[J]. IEEE Access, 2017, 5: 19269-19281 doi: 10.1109/ACCESS.2017.2754447 [9] FANG H Z, ZOU K X, YI D W, et al. The study of spacecraft telemetry data prediction based-on SERTS model[C]//Proceedings of 2011 Prognostics and System Health Management Conference. Shenzhen, China: IEEE, 2011: 1-5 [10] 石梦鑫. 基于注意力机制神经网络的遥测数据预测方法[D]. 北京: 中国科学院大学(中国科学院国家空间科学中心), 2020SHI Mengxin. Telemetry Data Prediction Method Based on Attention Mechanism Neural Network[D]. Beijing: The University of Chinese Academy of Science (National Space Science Center, Chinese Academy of Science), 2020 [11] AHMED A, ALY M, GONZALEZ J, et al. Scalable inference in latent variable models[C]//Proceedings of the 5 th ACM International Conference on Web Search and Data Mining. Seattle, Washington, USA: ACM, 2012 [12] HWANG Y, TONG A, CHOI J. Automatic construction of nonparametric relational regression models for multiple time series[C//Proceedings of the 33 rd International Conference on International Conference on Machine Learning. New York, NY, USA: JMLR. org, 2016 [13] JING H, SMOLA A J. Neural survival recommender[C]//Proceedings of the 10 th ACM International Conference on Web Search and Data Mining. Cambridge, United Kingdom: ACM, 2017 [14] LAI G K, CHANG W C, YANG Y M, et al. Modeling long- and short-term temporal patterns with deep neural networks[C]//Proceedings of the 41 st International ACM SIGIR Conference on Research & Development in Information Retrieval. Ann Arbor, MI, USA: Association for Computing Machinery, 2018: 95-104. DOI: 10.1145/3209978.3210006 -
-
下载:
计量
- 文章访问数: 699
- HTML全文浏览量: 162
- PDF下载量: 64
-
被引次数:
0(来源:Crossref)
0(来源:其他)
