基于动态加权集成学习的遥测数据预测方法

A Telemetry Data Prediction Method Through

摘要:

飞行任务中的遥测数据是飞行器中各子系统监测模块顺序产生的多维时间序列，其反应各子系统功能是否正常，对其精准预测是研判的重要依据。针对已有时间序列预测算法会随时间劣化的缺点，提出基于集成学习原理的动态加权神经网络集成算法，通过神经网络强数据拟合能力，集成学习算法具有的泛化特性和动态加权算法适应数据的漂移变化特性，提升算法的整体预测精度。实验表明该算法对预测精度提高效果显著，一定程度抑制数据的漂移。

Abstract:

Telemetry data of flight task are multi-dimensional time-series data streams sequentially produced by subsystem, It reflects whether the function of each subsystem is normal, Accurate prediction is an important basis for research and judgment. Aiming at the disadvantage that the existing time series prediction algorithm will deteriorate over time, A dynamic weighted neural network integration algorithm based on ensemble learning principle is proposed. By means of Strong data fitting ability of neural network, generalization characteristic of The ensemble learning algorithm, has the and adaptability of the dynamic weighting algorithm for the drifting characteristic of data, The overall prediction accuracy of the algorithm is improved, Experiments show that this algorithm can improve the prediction accuracy remarkably， and Restrain data drift to some extent.