基于混合降维森林的天线跟踪数据实时处理

Hybrid Dimensionality Reduction Forest for Real-Time Antenna Tracking Data

随着卫星数量和星上动作快速增加,卫星测控站测控任务也随之指数式增加,导致传统的24小时人工监视任务难以实现。基于数据挖掘的大数据实时分析技术能有效解决该问题,但面临样本维度高和标记数据不平衡的问题。提出一种新的集成学习模型——混合降维森林(HDRF),来处理复杂的天线跟踪数据。首先通过提出的样本特征化转换过程(SFTP)将异常样本转化为拓展的样本特征,其次通过基于树的特征选择算法挑选样本的强特征,并使用主成分分析(PCA)算法对新特征和未选择特征进行降维,生成具有有效性和补偿性的新特征。最后,在5个真实数据集上的实验表明,提出的算法优于其他主流的集成分类算法,且对天线跟踪数据实时处理切实有效。

With the rapid increase in the number of satellites and its movements,the TT&C tasks of the measurement and control station for satellites also increase exponentially,making it impossible to achieve 24-hour manual monitoring.The real-time analysis technology of big data based on data mining can solve this problem effectively.However,these methods have two challenges:the dimension of the data sets is very high and training data is unbalanced.To address the above limitations,we propose a novel ensemble learning model named hybrid dimensionality reduction forest(HDRF)to process complex antenna tracking data.First,a sample-feature based transformation process(SFTP)is proposed to generate the extended features from abnormal samples.Then,a tree-based feature selection algorithm is employed to partition effective features,and principal component analysis(PCA)is applied to compress the unselected features and the extended features into the new features which are compact and compensatory.Experimental results on 5 high-dimensional real data sets verify that our method outperforms mainstream classifier ensemble methods.Further,the realtime processing of antenna tracking data is effective.