基于决策树的卫星故障诊断知识挖掘方法

摘要

According to the problems of difficulty to get prior knowledge and precision decreasing of threshold based diagnosis since the satellite function and performance degeneration. We have presented decision tree based satellite Diagnosis Knowledge Mining Method using a lot of fault cases and abnormity data which were from the satellite in-orbit management. In this method it uses the max gain ratio attributes to be the partition attribute,using which mining the best partition point to rebuild the threshold value. It uses the pruning strategy to avoid the decision tree fitting to large. Finally it creates the fault diagnose knowledge. We have used the method to mine extract rules in massive real data and fault cases of in-orbit satellite. We have proved that the proposed method is efficient to extract satellite fault diagnosis rules via extracting the rules from the real cases for increasing the diagnosis accurate and decreasing the error rate.%针对目前卫星在轨故障诊断后验证知识获取困难,随着卫星在轨运行功能或性能退化导致门限诊断精度下降的问题,本文深入研究了卫星在轨管理过程中积累的异常数据和故障案例,提出了一种基于决策树的在轨卫星故障诊断知识挖掘方法.该方法选择信息增益率最大的属性作为分割属性,通过挖掘数据获取各属性的最优分割点建立门限,利用剪枝策略防止决策树过拟合或深度过大,最后梳理决策树生成故障诊断知识.通过对算例和对实际在轨数据进行挖掘,提取故障诊断知识,对知识的精度和差错分析,结果说明了本文所用方法能够提高知识的准确性同时降低误警率.