集成LSTM的航天器遥测数据异常检测方法

航天器作为一类集合结构、热控、电源、姿轨控等的复杂系统,遥测数据是地面判断其在轨性能的唯一依据,而有效的异常检测是保证航天器在轨可靠运行的基础要素。针对遥测数据连续、离散样本混合且样本变化高度关联于指令的数据异常检测问题,提出一种基于集成长短期记忆网络(LSTM)的航天器遥测数据异常检测方法。利用LSTM强大的非线性建模能力,结合矩阵范数实现对遥控指令的多模式挖掘,并通过多LSTM预测模型的构建以及有效集成,提升模型对于航天器复杂工况的适应性,进而有效标记遥测数据中的异常。通过对NASA公布的2个类型航天器的遥测数据进行实验验证,结果表明,与基于LSTM的遥测数据异常检测方法相比,所提出的方法异常检测率提升明显,尤其适合检测上下文类型异常。测试结果验证了方法的可行性,可为航天器地面运控提供有效的数据判读能力。

Spacecraft telemetry data anomaly detection method based on ensemble LSTM

Spacecraft is a kind of extraordinary complex systems consisting of integrating structure, thermal control, power, attitude, orbit, and so on. Telemetry data is the only basis to judge the on orbit spacecraft performance on ground. The effective anomaly detection is a fundamental element to ensure the reliable operation of the spacecraft. In this paper, aiming at the data anomaly detection problems that the telemetry data are the mixture of continuous and discrete samples, and the sample variation is highly correlated with the instructions, a spacecraft telemetry data anomaly detection method is proposed based on ensemble long short term memory (LSTM) network. The strong nonlinear modeling ability of LSTM is utilized; with matrix norm, the multiple mode mining of tele control instruction is achieved; through the construction and effective ensemble of the multiple LSTM prediction model, the adaptability of the model to the complicated spacecraft operating condition is improved; then, the anomaly in the telemetry data is effectively labeled. The telemetry data of two kinds of spacecraft from NASA are detected in experiment. The result indicates that the anomaly detection rate of the proposed method is promoted obviously compared with the telemetry data anomaly detection method based on LSTM, the proposed method is especially suitable for contextual anomaly discovery. The test results verify the feasibility of the proposed method, and the study provides an effective data interpretation ability for the ground operation and control of spacecraft.