遥测参数数据载荷状态判别集成学习方法

针对载荷单机设备遥测参数维度高、数据量大、存在类别不平衡、无法直观判别单机设备运行情况等问题,考虑到航天任务对可解释性的要求,提出一种基于信息增益参数特征选择和集成学习方法的载荷单机状态快速识别方法。采用统计量性质和信息增益子集搜索方法对遥测数据进行特征筛选降维,通过集成学习模型算法实现载荷单机设备状态的自适应识别分类。所提方法将信息增益的参数分类信息量评价准则和集成学习拟合能力强、类别不平衡下准确率高和抗噪能力强等优点相结合,兼顾模型特征和结果的可解释性,提供了重点参数发现功能。采用科学卫星任务真实载荷遥测参数数据对该方法进行了验证,整体识别准确率高于90%,少数样本亦可准确识别,整体效果可达到在轨任务要求,证明了所提方法的有效性和实用性。

Ensemble learning for state recognition of payload from telemetry data

In order to deal with various complex telemetry problems, such as high dimensionality, huge volume, unbalanced categories and failure to intuitively make sense about states of payload, and considering the requirement of interpretability in space mission, a general method for fast identification of payload based on information gain and integrated learning method was proposed. Sample statistics and information gain was used to select features and reduce the dimension of the telemetry data; meanwhile, the integrated learning algorithm was used to complete the adaptive recognition and classification about payload states. The proposed method combined the advantages of the parameter classification information evaluation criteria of the information gain and strong modeling, high accuracy and strong anti-noise ability under unbalanced category samples. Furthermore, the model had to possess the property of being explanatory and able to find the key parameters. The method was verified by experiments using actual mission data, which was tested using the payload telemetry data on operational scientific satellite mission. Following that, an state-of-art result, of which the overall recognition accuracy is higher than 90 percent and a few samples can also be identified, covered mission requirement in all and proved the effectiveness and practicability.