基于时间序列的航天器遥测数据预测算法研究

发布时间：2018-09-07 15:42

【摘要】：进行科学决策的重要依据是正确的预测。由于航天器运行在复杂的空间环境中,为了更加高效的对在轨航天器进行管理,有必要对航天器的运行状态进行预测。而航天器的遥测参数变化趋势可以有效的反映其在空间环境中的运行状况。航天器遥测参数中包含了设备状态的详细信息,根据数据信息变化规律,可以为遥测参数变化状态建立合适的预测模型。基于时间序列的预测算法在航天器遥测数据研究领域的应用中前景广阔。时间序列数据中的参数值和历史参数具有显著的联系,历史参数的变化能够影响未来参数的变化趋势,体现出了参数具有记忆性。本文通过对当前常用的预测方法进行了简单介绍,分析总结了相关预测模型在处理线性数据中的优势以及在处理非线性数据时的不足。为了解决非线性数据处理时的不足,引入了具有非线性映射功能的人工神经网络。目前,BP网络的发展最为成熟。它对于解决非线性数据预测具有强大的优势。高效的非线性映射能力,是它的显著优势。它对于预测参数没有明显的要求,只要对历史遥测参数进行有效的学习,就能够对数据的未来变化进行预测。然而标准的BP神经网络预测模型本身也存在着一些缺点。针对算法存在的这些缺点,提出了相应的优化方法。实际中的遥测数据序列往往比较复杂,在特定的时间段里非线性关系和线性关系同时存在。因此,本文提出将基于时间序列的遥测数据相关性拆分成非线性模块和线性模块。由于时间序列具有可分解性,所以可以先通过线性时间序列AR模型把遥测数据中的线性主体部分进行预测。下一步将拆分后的非线性序列部分,通过BP算法处理。最终的输出由计算出的非线性部分和线性部分叠加组成。同时,由于遗传算法(GA)是一种全局优化的算法,针对BP网络存在的不足,结合使用GA算法来优化BP网络的初始权阈值,从而缓解BP网络容易陷入极小值的问题。本文将所构造的预测模型应用到对预测某航天器遥测数据变化趋势的实例中,经过多次仿真实验,得到的结果表明,该AR-BP-GA综合预测算法是符合要求的。并且实验仿真效果比只使用一种线性AR模型的结果要好。最终验证了我们所提出的综合预测算法更加实用和有效。
[Abstract]:The important basis for scientific decision-making is correct prediction. In order to manage the orbiting spacecraft more efficiently, it is necessary to predict the state of spacecraft operation because the spacecraft operates in a complex space environment. The change trend of spacecraft telemetry parameters can effectively reflect its operation in space environment. The spacecraft telemetry parameters contain the detailed information of the equipment state. According to the variation rule of the data information, a suitable prediction model can be established for the change state of the telemetry parameters. The prediction algorithm based on time series has a bright future in the research field of spacecraft telemetry data. The change of historical parameters can influence the change trend of future parameters, which shows that the parameters are memorized. In this paper, the common prediction methods are briefly introduced, and the advantages of correlation prediction model in dealing with linear data and the shortcomings in dealing with nonlinear data are analyzed and summarized. In order to solve the problem of nonlinear data processing, an artificial neural network with nonlinear mapping function is introduced. At present, the development of BP network is the most mature. It has a powerful advantage in solving nonlinear data prediction. Efficient nonlinear mapping capability is its significant advantage. It has no obvious requirement for the prediction parameters. As long as the historical telemetry parameters are effectively studied, the future changes of the data can be predicted. However, the standard BP neural network prediction model itself also has some shortcomings. Aiming at these shortcomings of the algorithm, a corresponding optimization method is proposed. In practice, telemetry data sequences are often more complicated, and both nonlinear and linear relationships exist in specific time periods. Therefore, in this paper, the correlation of telemetry data based on time series is divided into nonlinear module and linear module. Because time series are decomposable, the linear principal part of telemetry data can be predicted by linear time series AR model. The next step will split the nonlinear sequence part, through the BP algorithm. The final output consists of the calculated nonlinear part and the linear part superposition. At the same time, because genetic algorithm (GA) is a global optimization algorithm, the GA algorithm is used to optimize the initial weight threshold of BP network in order to alleviate the problem that BP network is easy to fall into the minimum value. In this paper, the constructed prediction model is applied to an example of predicting the trend of a spacecraft telemetry data. After many simulation experiments, the results show that the AR-BP-GA synthetic prediction algorithm meets the requirements. And the simulation result is better than that using only one linear AR model. Finally, it is proved that the proposed comprehensive prediction algorithm is more practical and effective.
【学位授予单位】：西安工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：V557;TP18

【参考文献】