Research on the Application of ARIMA-SVR Combination Model in Satellite Telemetry Parameter Prediction
-
摘要: 为辅助卫星在轨运行提供决策分析支持,结合卫星遥测参数的时间序列特性,利用一种ARIMA-SVR组合预测方法,通过对卫星遥测参数进行预测,判定实际遥测数据是否处于正常范围。该组合模型利用ARIMA模型对预处理后的数据进行线性拟合,并利用SVR模型对数据的非线性部分进行补偿。以KX09卫星星敏A的温度遥测数据为基础,分别利用组合模型对短期及中期星敏A温度进行预测,得出短期和中期均方根误差(RMSE)分别为0.768和0.968,相比单一ARIMA模型,短中期RMSE分别提高46.2%和16.4%。此外,对该卫星陀螺B的x轴角速度进行了短中期预测:短期预测中,组合模型比单一ARIMA模型的RMSE提高71.2%;中期预测中,组合模型比单一ARIMA模型的RMSE提高64.2%。实验结果表明,ARIMA-SVR组合模型为保证卫星在轨正常运行提供了有效的决策分析支持。
-
关键词:
- 卫星正常运行 /
- 遥测参数预测 /
- 时间序列 /
- ARIMA-SVR组合模型
Abstract: In order to provide decision analysis support for assisting the in-orbit operation of satellite, combining with the time series characteristics of satellite telemetry parameters, an ARIMA-SVR combination prediction method is used to judge whether the actual telemetry data is in the normal range through the prediction of satellite telemetry parameters. In this model, ARIMA model is used to linearly fit the preprocessed data, and SVR model is used to compensate the nonlinear part of the data. Based on the telemetry data of temperature of star sensor A in KX09 satellite, the combination model was used to predict the short-term and medium-term temperature of star sensor A, and the Root Mean Square Error (RMSE) of the short-term and medium-term temperature of star sensor A was 0.768 and 0.968, respectively. Compared with the single ARIMA model, the RMSE of the short and medium-term temperature of star sensor A was 46.2% and 16.4% higher than that of the single ARIMA model respectively. In addition, the x-axis angular velocity of gyro B on the satellite is predicted in short and medium term. In the short term prediction, the RMSE of the combined model is increased by 71.2% compared with that of the single ARIMA model. In the medium prediction, the RMSE of the combination model is 64.2% higher than that of the single ARIMA model. Experimental results show that the ARIMA-SVR combination model provides effective decision analysis support for ensuring the healthy in-orbit operation of satellites. -
表 1 三种核函数的RMSE值对比
Table 1. Comparison of RMSE of three kernel functions
核函数 RMSE RBF 0.936 Linear 1.532 Poly 1.533 表 2 PSO算法参数设置
Table 2. Parameter setting of PSO algorithm
参数名称 数值 粒子数量 100 粒子维度 2 最大迭代次数 500 局部学习因子 2 全局学习因子 2 惯性因子 0.8 参数最大值 15 参数最小值 0.001 表 3 温度预测结果RMSE统计
Table 3. RMSE statistics of prediction results of temperature
模型 RMSE 短期 中期 ARIMA 1.428 1.159 ARIMA-SVR 0.768 0.968 表 4 角速度预测结果RMSE统计
Table 4. RMSE statistics of prediction results of angular velocity
模型 RMSE 短期 中期 ARIMA 2.708 3.181 ARIMA-SVR 0.779 1.138 -
[1] 任国恒. 同步卫星遥测数据相关性分析与研究[D]. 西安: 西安工业大学, 2011REN Guoheng. Correlation Analysis and Research on the Telemetry Data of Synchronous Satellites[D]. Xi’an: Xi’an Technological University, 2011 [2] 余文艳. 机器学习在有效载荷PHM系统中的应用研究[D]. 北京: 中国科学院大学(中国科学院国家空间科学中心), 2018YU Wenyan. Application and Research of Machine Learning in Payload PHM System[D]. Beijing: University of Chinese Academy of Sciences (National Space Science Center, the Chinese Academy of Sciences), 2018 [3] ZHANG G P, PATUWO B E, HU M Y. A simulation study of artificial neural networks for nonlinear time-series forecasting[J]. Computers & Operations Research, 2001, 28(4): 381-396 [4] DONATE J P, LI X D, SÁNCHEZ G G, et al. Time series forecasting by evolving artificial neural networks with genetic algorithms, differential evolution and estimation of distribution algorithm[J]. Neural Computing and Applications, 2013, 22(1): 11-20 doi: 10.1007/s00521-011-0741-0 [5] 朱俊鹏, 赵洪利, 杜鑫, 等. 长短时记忆神经网络在卫星轨道预报中的研究[J]. 兵器装备工程学报, 2017, 38(10): 127-132 doi: 10.11809/scbgxb2017.10.026ZHU Junpeng, ZHAO Hongli, DU Xin, et al. Application of long short-term memory neural network to orbit prediction of satellite[J]. Journal of Ordnance Equipment Engineering, 2017, 38(10): 127-132 doi: 10.11809/scbgxb2017.10.026 [6] 任国恒, 朱变, 朱海. 马特拉算法在遥测数据短期预测中的应用[J]. 武汉工程大学学报, 2014, 36(2): 73-78 doi: 10.3969/j.issn.1674-2869.2014.02.014REN Guoheng, ZHU Bian, ZHU Hai. Application of Mallat algorithm in short-term forecasting of telemetry data[J]. Journal of Wuhan Institute of Technology, 2014, 36(2): 73-78 doi: 10.3969/j.issn.1674-2869.2014.02.014 [7] 刘家庆, 张弘鹏, 郭希海, 等. 基于SVR残差修正的光伏发电功率预测模型[J]. 电力工程技术, 2020, 39(5): 146-151LIU Jiaqing, ZHANG Hongpeng, GUO Xihai, et al. Prediction model of photovoltaic power generation based on SVR residual correction[J]. Electric Power Engineering Technology, 2020, 39(5): 146-151 [8] 黄红梅. 应用时间序列分析[M]. 北京: 清华大学出版社, 2016HUANG Hongmei. Applied Time Series Analysis[M]. Beijing: Tsinghua University Press, 2016 [9] 张森, 石为人, 石欣, 等. 基于偏最小二乘回归和SVM的水质预测[J]. 计算机工程与应用, 2015, 51(15): 249-254 doi: 10.3778/j.issn.1002-8331.1308-0117ZHANG Sen, SHI Weiren, SHI Xin, et al. Water quality prediction based on partial least squares and Support Vector Machine[J]. Computer Engineering and Applications, 2015, 51(15): 249-254 doi: 10.3778/j.issn.1002-8331.1308-0117 [10] LUO Xueke, HE Yunxiao, LIU Peng, et al. Water quality prediction using an ARIMA-SVR hybrid model[J]. Journal of Yangtze River Scientific Research Institute, 2020, 37(10): 21-27 doi: 10.11988/ckyyb.201908087 [11] 周志华. 机器学习[M]. 北京: 清华大学出版社, 2016ZHOU Zhihua. Machine Learning[M]. Beijing: Tsinghua University Press, 2016 [12] 向超. 基于ARIMA-SVR组合模型的动力煤价格预测与实证研究[D]. 北京: 对外经济贸易大学, 2019XIANG Chao. Prediction and Research of Steam Coal Price Based on ARIMA and SVR Model[D]. Beijing: University of International Business and Economics, 2019